Source: TENNESSEE STATE UNIVERSITY submitted to
IMPROVING PRODUCTION AND PROFITABILITY OF NURSERY PRODUCTION USING UNMANNED AIRCRAFT SYSTEM
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
OTHER GRANTS
Reporting Frequency
Annual
Accession No.
1018055
Grant No.
2019-38821-29062
Cumulative Award Amt.
$599,925.00
Proposal No.
2018-04961
Multistate No.
(N/A)
Project Start Date
Jan 1, 2019
Project End Date
Dec 31, 2023
Grant Year
2019
Program Code
[EQ]- Research Project
Recipient Organization
TENNESSEE STATE UNIVERSITY
3500 JOHN A. MERRITT BLVD
NASHVILLE,TN 37209
Performing Department
(N/A)
Non Technical Summary
This is a standardintegrated (research and extension)project for the 1890 CBG Program. A major concern of nursery producers is timely and effective management ofdiseases and abiotic stresses. Another concern is the availability and escalating costs of labor.The goal of this project is todesign, optimize and implement an unmanned aircraft system (UAS) that can operate in nursery systems to improvethe production and profitability of nursery crops by 1) early detection of foliar diseases and abiotic stresses, and 2) reduce the cost of labor for determination of individual tree criteria for sale. The specific objectives of this proposal are: 1) design/build an UAS to operate within a nursery production environment, 2) evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements, 3)assess the economic impact of these technologies on nursery production costs and profitability,and 4)technology and information transfer to practitioners.This projectwill result in an UAS to improve nursery production,with the capability to move above the plant canopy, between rows of plants, and hover directly beside a plant to collectfoliar disease, abiotic stressand plant measurement datawithin a nursery production environment. Practitioners and students will have access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business.
Animal Health Component
50%
Research Effort Categories
Basic
20%
Applied
50%
Developmental
30%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
2162110116040%
4022110202030%
2052110102020%
6012110301010%
Knowledge Area
601 - Economics of Agricultural Production and Farm Management; 205 - Plant Management Systems; 216 - Integrated Pest Management Systems; 402 - Engineering Systems and Equipment;

Subject Of Investigation
2110 - Ornamental trees and shrubs;

Field Of Science
2020 - Engineering; 1160 - Pathology; 3010 - Economics; 1020 - Physiology;
Goals / Objectives
Nursery crops are produced with an annual sale value of $5.1 billion in the United States.Nursery crops in Tennessee account for asignificant amount of agricultural production with annual sales of $132 million (USDA, 2014). Nursery producers are vulnerable to losses due to plant diseases and, in 2013;these losses were estimated to be approximately 15% of the farm gate value for field and container nursery production (Williams-Woodward, 2013). Although fungicides/bactericides are available for nursery production, they are most effective in reducing disease severity/incidence when applied during the early stage of disease development. Traditionally, nursery producers follow a calendar-based program for making fungicide/bactericide applications or conduct visual inspections for deciding when to start fungicide/bactericide applications. Nursery plants vary widely in size and leaf architecture, and different diseases cause different types of symptoms, therefore only specially trained personnel are able to perform accurate disease scouting.Plant growth measurements of nursery crops (such as height, width, stem diameter, etc.) dictate when a plant is at a marketable stage. These measurements must also be performed manually by nursery personnel. This dependence on physical labor is a critical weakness for the nursery industry for two primary reasons: 1) labor is costly, cutting into industry profits and increasing prices for customers; 2) the agricultural labor supply is diminishing due to potential immigration reforms and demographic changes in Mexico (Charlton and Taylor, 2016). Thus,developing unmanned aerial systems (UAS) applications for early detection and identification of diseases and abiotic stresses and quality measurements will help ensuresustainability for the nursery industry. Since fungicides/bactericides are expensive and there is high risk of fungicide/bactericide resistance development with calendar-based pesticide applications due to over-use, this new tool will help to reduce the risk of resistance development and the overall costs of pesticide applications. These tools will also help us to determine spatial distribution of foliar diseases and how this distribution affects crop growth and quality.Large- and medium-scale nursery producers have consulted with us on the need to develop and validate effective alternative tools such as UAS for early detection and identification of foliar diseases and abiotic stresses and for quality measurements strategies. Several nursery producers, theTennessee State University OtisL.Floyd Nursery Research Center(TSUNRC) advisory board, the Tennessee Department of Agriculture (TDA) plant pathologist, and TDA plant inspectors have participated in the development of this integrated (research and Extension) proposal by telephone, e-mail and group discussion at the TSUNRC. Stakeholder support for the proposed work is further indicated by the attached Letters of Support.The long-term goal of this proposed integrated project is todesign, optimize and implement an UAS that can operate in woody ornamental nursery systems to improvethe production and profitability of nursery crops by 1) early detection and identification of foliar diseases and abiotic stresses, and 2) reduce the cost of labor for determination of individual tree criteria for sale such as plant size and quality.The specific objectives of this proposal are: 1) design/build an UAS to operate within a nursery production environment, 2) evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors (water, nutrients, etc.), and plant growth measurements (height, width, stem diameter, etc.), 3)assess the economic impact of these technologies on nursery production costs and profitability,and 4)technology and information transfer to practitioners[nursery producers, industry professionals and service providers (possible new small business opportunities)].?
Project Methods
Objective 1. Design/build an UAS to operate within a nursery production environmentNursery crops are produced with a high density of plants within a relatively small area.An UAS will be designed with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collectfoliar disease, abiotic stress and plant measurement data. The UAS will be equipped with combination of cameras, other sensors (thermal, multispectral, hyperspectral and/or time-of-flight) and a data mining station. ADJI Phantom 4 Professional+ Quadcopter will be used for the unmanned aerial vehicle (UAV) part of this UAS. The UAS will be composed of a UAV, which is equipped with GPRS signal receiver,Inertial Measurement Unit (IMU),a gripper, a robotic manipulator, a depth camera system, sensors and a laptop with a database for recording the data and data mining. The gripper will be customized and attached to a robotic arm manipulator on the UAV to collect leaf samples inside the branches. To determine the shape and sizes of the trees, a depth camera system-a stereo camera system and time-of-flight sensors, will be used to generate a 3D mesh model of the trees separately as the UAV flies around these trees and uploads data to the station. The GPRS system will be used to localize the UAV and the trees. These components will be evaluated to minimize the weight of the UAV and to maximize data collection accuracy for the UAS.Objective 2.Evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements2.1. Early detection and identification of foliar diseasesIn Years 1 and 2, a UAS equipped with a depth camera, time-of-flight sensor, GPRS, IMU as well as other sensor such asthermal, multispectral or hyperspectralwill be evaluated for its effectiveness in detecting foliar diseases compared with visual scouting techniques.Model nursery crops will be flowering dogwood plants produced indifferent nursery production platforms (field, pot-in-pot, and containeron gravel and black fabric ground cover).The experimental design will be arandomized complete block designwith a minimum of 25 plants per setting.An UAS will be used daily to determine natural occurrence of dogwood powdery mildew and anthracnose symptoms during the early production season and sampling will be done using a gripper built on the UAS. Disease severity and progress (AUDPC)will be monitored to determine the sensitivity of the UAS camera and sensors. The visual scouting will be examined by a trained personnel/researcher using a0-100% scale for foliage area affected. Collected samples from plots will be examined for identification bya trained personnelin the plant pathology laboratory.Labor and time spent assessing disease severity and positive confirmation timing will be recorded for comparison.In Year 3, several other studies will be conducted withsame environment and experimental design as described before to determine the accuracy and consistency of the UAS for early detection of foliar diseases on other woody ornamental plants with various color and shape of leaves and making comparisons with visually evaluated plants.2.2. Determination of abiotic stressesNursery crops require adequate water and nutrients for healthy growth, yet symptoms for drought or nutrient stress may not be visible until after the stress has significantly progressed. Corrective actions can be taken, but it may take weeks or months for the crop to recover, if at all. Sensors are available that can detect drought stress and leaf greenness in plants prior to when more obvious visible symptoms of stress occur. An UAS equipped with these sensors would be a valuable tool for detecting signs of drought and nutrient stress in a timely manner so that producers could take corrective actions. Several studies will be conducted to evaluate UAS-mounted sensors for detecting drought stress at various stages using different irrigation regimes with two levels of water deficit compared with visual ratings collected by researchers in Years 1 and 2. Plants will be evaluated at different levels of water stress to compare the sensitivity of the UAS with the visual ratings. In a separate study, UAS sensors will be evaluated for ability to detect changes in leaf greenness based on different nutrient deficiencies and compared with visual ratings and readings from a handheld leaf greenness meter. Plants will be grown with three different levels of nitrogen, phosphorus and potassium to produce plants with varying levels of chlorosis for use to compare the accuracy of detection between the UAS and manual readings. The model nursery crop for these studies will be dogwood plant produced withsame environment. The experimental design will be asplit plotwith a minimum of 25 plants per setting on a randomized complete block design, control plants will be well-watered and fertilized. In Year 3, several other studies will be conducted to determine the accuracy and consistency of the UAS for determination of abiotic stresses on other plants with various color and shape of leaves and making comparisons with visually evaluated plants.2.3. Plant growth measurementNursery producers must manually measure each plant to grade the plant size and quality for sale. An UAS fitted with sensors to take plant measurements would allow producers to automate this process and allocate labor to other tasks. An UAS equipped with a camera and/ortime-of-flight sensor will be evaluated for measuring plant height, width, and stem diameter in Years 1, 2 and 3. The studies will be conducted on plants with different morphological characteristics to determine the accuracy and consistency of the measurements determined by the UAS and making comparisons with manually measured plants.Objective 3.Assess the economic impact of these technologies on nursery production costs and profitabilityThe use of an UAS in nursery production systems will allow nursery producers to automate certain tasks such as scouting for disease and abiotic stress issues and measuring/monitoring crop growth. Although a technically trained person will be required to operate the UAS and manage the data, producers may be able to operate with fewer employees overall. In Years 2 and 3, we will conduct several studies to compare the time required to conduct manual scouting/measuring with how much time is saved by utilizing the UAS.Objective 4.Technology and information transfer to practitionersOnce the UAS has been developed and optimized for use in nursery production systems, we will conduct workshops for practitioners detailing the advantages of using UAS, rules and regulations of using UAS, and research updates in Years 2 and 3 at the TSUNRC. We will also conduct training sessions/demonstration field days for practitioners to teach how to operate the UAS, how to download and interpret the data and how to develop business plan in Years 2 and 3 at the TSUNRC auditorium and demonstration plots. Questionnaires will be presented to participants following each activity session to identify the impact of the activity and suggestions for changes which will be used to revise activity for the next session. Three months after each activity, a general assessment will be delivered to participants to determine their decision-making in adoption of an UAS on their farms or creating a small business.Our outreach efforts will also include dissemination of information to the nursery industry through local crop management and nursery conferences as well as national conferences, meetings, written materials, video clips, social media.

Progress 01/01/19 to 12/31/23

Outputs
Target Audience: This integrated project resulted in an UAS to improve nursery production, with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collect foliar disease, abiotic stress and plant measurement data within a nursery production environment. Nursery producers, industry professionals as well as undergraduate students had access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension and research products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided? Krishna Neupane was able to learn experimental design, data collection, tabulation and analysis. Research associate was able to analyse the survey results. They were able to attend the society meetings to attend to present their results. How have the results been disseminated to communities of interest? The products, results and impacts of this integrated project were disseminated through a number of methods and to a number of target audiences. Presentations. The project team was involved in developing workshops, training sessions/demonstration field days, research and Extension symposium/meeting presentations.Publications. fact sheets, Extension articles and technical bulletins were developed and the target audiences were nursery producers, Extension agents, students, scientists and other stakeholders. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Two trials (2021 and 2022) were conducted to evaluate the physiological changes induced due to drought conditions. In an outdoor setting, trees were organized in a randomized complete block design. Three different irrigation treatments were applied at 125%, 25% and 10% (control, moderate and severe drought, respectively) of their daily water usage (evapotranspiration). The physiological parameters Normalized Difference Vegetation Index (NDVI) and leaf moisture potential (PMS) were collected every week for one month. Plant growth data (height and width) were collected at the beginning and the end of the study. NDVI data collected with a handheld NDVI meter and a Sentera NDVI sensor mounted on a UAV were correlated for ground truthing. In 2021, control plants had greater plant width increase and shoot biomass whereas no significant differences in growth were observed among the treatments in 2022. In both trials, NDVI was the greatest for control plants compared to the other treatments on the 7th, 14th, 21st and 27th days. In both studies, no differences were observed for PMS on day 7 but was greatest for controls on the 14th, 21st and 27th days. The correlation between the handheld NDVI and UAV NDVI was found to be strong and positive ranging from 0.84 to 0.93 (Trial 1:P≤ 0.0001,P≤ 0.0001,P= 0.0002 andP≤ 0.0001; Trial 2:P= 0.0002,P≤ 0.0001,P≤ 0.0001 andP≤ 0.0001 for week 1 to week 4, respectively). This information will be applicable to understanding the physiology of the crop and the inclusion of emerging technology in crop production and monitoring. Early season monitoring of nutrient stress is important in nursery crops to optimize management practices and ensure healthy crop production. Two different irrigation systems (drip and overhead irrigation) were used in the study. Two rates (low and high) of controlled-release fertilizer were used with no fertilizer as a control treatment. Data were recorded for plant height, stem diameter, substrate pH and electrical conductivity (EC), chlorophyll content, normalized difference vegetation index (NDVI), visual observation of plant quality, and leaf nutrient content.In the results, increase in plant height and stem diameter was greater among the fertilized maple tree whereas no differences were observed in the flowering dogwoods for increase in plant height. NDVI was greater for drip irrigation for both fertilizer rates in both red maples and flowering dogwoods. A positive correlation of 73% to 83% was observed for red maples and 79% to 83% for flowering dogwoods between handheld NDVI and unmanned aerial vehicle mounted NDVI sensor. Greater chlorophyll content was observed in both rates of fertilizer in both tree species. In red maple, high fertilizer rate resulted in greater substrate pH with overhead irrigation whereas substrate EC was greater among red maples with drip irrigation. In flowering dogwood, no differences were observed. Varied response was observed among the treatments for nutrient content, however both rates of fertilizer application were sufficient for both tree species. Drip irrigated red maples had higher nitrogen and phosphorous content whereas nitrogen content was higher in both irrigation systems in flowering dogwoods. This study provides useful insights to understand the effect of nutrient stress on tree growth and application of sensing technology for monitoring and early detection of nutrient stress in container-grown nursery crops. ?Early and accurate detection of diseases and implementation of efficient disease management practices are crucial in reducing the economic impact associated with plant disease outbreaks.Based on survey responses from dogwood nursery growers in Tennessee, scouting was identified an important disease management practice adopted by a majority of growers for disease management in field-grown, container-grown, and pot-in-pot production systems. Our results showed a significantly positive correlation between disease severity and scouting frequency for dogwood plants grown in container and pot-in-pot production systems. Our efficiency measure is a self-rated efficacy scale perceived by the nursery growers about their existing disease management system in nursery plants. Significantly positive correlation was found between the efficacy of disease management and the number of workers involved in scouting and positive associations were shown by the worker hours in scouting with the grower's experience to other dogwood disease detection. The majority of nursery growers followed a set spray schedule between May and October with applications scheduled every other week. Additionally, our results showed significantly positive correlations between the efficacy and spray related factors, such as, disease severity and worker hours spent in spraying, the efficacy of disease management and spraying frequency in field-grown dogwoods, and the foliar spray costs and the efficacy of disease management. We estimated around $379 per acre per year average costs for dogwood disease management, which the growers find one of the major components of the dogwood production budget. Moving to automated systems of disease scouting and management has the potential to reduce the cost of these labor-intensive disease management practices of dogwood production.

Publications

  • Type: Journal Articles Status: Accepted Year Published: 2024 Citation: Liyanage, K.H.E., Khanal, A., Witcher, A., Liyanapathiranage, P., and Baysal-Gurel, F. 2024. Assessing the Impact of Integrated Dogwood Disease Management Practices on Labor Needs and Production Costs in Tennessee Nurseries. HORTSCIENCE _ https://doi.org/10.21273/HORTSCI17717-24. (in press)
  • Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Neupane, K., Witcher, A., Delay, G. & Baysal-Gurel, F. 2023. Evaluation of effect of fertilizer rate on tree growth and detection of nutrient stress in different irrigation systems. HortScience. (under review)
  • Type: Journal Articles Status: Published Year Published: 2023 Citation: Neupane, K., Witcher, A., & Baysal-Gurel, F. 2023. Evaluation of physiological changes in flowering dogwood under drought conditions in a container production system. HortScience, 58(9), 1077-1084 https://doi.org/10.21273/HORTSCI17279-23
  • Type: Conference Papers and Presentations Status: Published Year Published: 2024 Citation: Liyanage, K.H.E., Witcher, A., Khanal, A., Baysal-Gurel, F. 2024. Investigating the Economic Impact of Dogwood Disease Management in Tennessee Nursery Industry. The 2024 annual meeting of the Southern Agricultural Economics Association. February 3-6, 2024. Atlanta, GA.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Hikkaduwa Epa Liyanage, K., Witcher, A., Khanal, A., and Baysal-Gurel, F. 2023. Assessing the disease management practices of dogwood production in the Tennessee nursery industry. 2023 TAS Meeting. Memphis, TN. November 18, 2023. (Poster presentation).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2023 Citation: Parajuli, M., Oksel, C., Neupane, K., Baysal-Gurel, F. 2023. Effect of plant defense elicitor in preventing ambrosia beetle attacks in flowering dogwoods exposed to simulated flood stress condition. Conference on Water, Climate, and Food Security. Prairie View, TX. March 6-10, 2023. (Poster presentation).


Progress 01/01/22 to 12/31/22

Outputs
Target Audience: This integrated project esulted in an UAS to improve nursery production, with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collect foliar disease, abiotic stress and plant measurement data within a nursery production environment. Nursery producers, industry professionals as well as undergraduate students had access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension and research products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided? Krishna Neupane was able to finish internship with BASF for 10 weeks- Summer 2022. Student learnt experimental design, data collection, tabulation and analysis. How have the results been disseminated to communities of interest? The products, results and impacts of this integrated project were disseminated through a number of methods and to a number of target audiences. Presentations. The project team was involved in developing workshops, training sessions/demonstration field days, research and Extension symposium/meeting presentations. Publications. eXtension publications, fact sheets, Extension articles and technical bulletins were developed and the target audiences were nursery producers, Extension agents, students, scientists and other stakeholders. What do you plan to do during the next reporting period to accomplish the goals? Objective 1. Design/build an UAS to operate within a nursery production environment. This research is being conducted using an interactive process approach, which needs continuous improvement of the design, concept, hardware, and software systems. The repeated trials and evaluations help us to get closer to our final prototype. The remaining final iterations and the development are summarized in the next sections. (1) Hardware The gripper design will be finalized and tested on collecting dogwood tree leaves in the nursery. (2) Software The user application will be finalized and tested with the users on the workstation. The database system and the data collection system will be optimized for faster data processing and gathering. The leaf detection system and the disease detection system will be improved with more data from the field. (3) Evaluations Final evaluations will be made and analyzed. The performance of the overall system will be recorded and analyzed in terms oftime, speed, correctness of measurements and efficiency. Objective 2. Evaluate the UAS for early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements Nutrient stress experiment will be conducted in spring 2023 for potential detection using UAV. Objective 3. Assess the economic impact of these technologies on nursery production costs and profitability Surveydata will be analyzed. Objective 4. Technology and information transfer to practitioners Outreach activities will be conducted.

Impacts
What was accomplished under these goals? Objective 1: 1.Hardware The main structure of the UAV system has been finalized and the appropriate hardware has been selected. The list of the hardware for the project is listed below: 1. ZED Depth Sensor 2K Stereo Camera 2. Mavic 2 Pro Drone 3. NVIDIA Jetson Nano 4. Auxiliary and battery setups 2. Software- Stereo Camera Control via Network On a new computing environment called Internet of Things (IoT) or Smart Object networks, a lot of constrained devices are connected to the Internet. The devices interact with each other through the network and provide new opportunities to the researchers. Cameras connected IoT devices are very popular and stereo cameras are one of them. The main problem on these devices, they require a lot of computation power, which is not possible in IoT devices. Instead of performing visual computational issues on these small devices, transferring the data taken by the camera into a high computational capable computer is one of the common approaches. In this objective, a stereo camera control application, which is managed via network is developed. The application can control camera, activating it, streaming visual output, and transferring the final recording 3D model generation. 2.1.Qt Framework and Editor (QtCreator) Qt Framework and Editor was used to design a user-friendly application to connect to our UAV system. Qt is the complete software development framework, which contains a set of highly easy to use and modularized C++ library classes and is loaded with APIs to simplify the application development. Qt produces highly readable, easily maintainable, and reusable code with high runtime performance and small footprint and it's cross-platform. Qt Creator is the main development integrated development environment (IDE) of the Qt framework. 2.2.Zed Camera The ZED is a passive stereo camera that reproduces the way human vision works. Using its two stereo cameras, the ZED creates a three-dimensional map of the scene by comparing the displacement of pixels between the left and right images. Up until now, 3D sensors have been limited up to perceiving depth at short range and indoors. The ZED stereo camera is the first sensor to introduce indoor and outdoor long range depth perception along with 3D motion tracking capabilities, enabling new applications in many industries: AR/VR, drones, robotics, retail, visual inspection and more. The ZED captures two synchronized left and right videos of a scene and outputs a full resolution side-by-side color video on USB 3.0. This color video is used by the ZED software on the host machine to create a depth map of the scene, track the camera position and build a 3D map of the area.Jetson Nano computer system and the developed communication application is used to turn the ZED into an IOT camera.? 3.Evaluation 3.1.Implementation We continue evaluating the hardware and software system to improve the performance of the UAV system. Following the completion of the data collection and image annotation stages, the labeled data was used as a training set for the proposed model for the purpose of classifying healthy and diseased leaves. A series of convolution layers are followed by BiFPN layer and then fully connected layers in the model. The convolutional layers are responsible for identifying features in the data, while the fully connected layers serve as classification layers. More accurate detectors for object detection, such as disease detection in leaves, have frequently been discovered to be more compute-demanding, which is not ideal, especially when the goal is to find more and more economical models which are efficient and requireminimal computing resources. EfficientDet, a Google Brain architecture, is built on top of EfficientNet , a convolutional neural network pre-trained for classification on the ImageNet image database. The Google Brain team developed a new family of detectors that are very efficient, accurate, and significantly faster. Several Convolution models have been developed and, for example, one of the models MobileNet V1 and MobileNet V2 gave us a way to implement a neural network, that is more computationally efficient. Implementing a computer vision algorithm for different types of devices like Jetson Nano and how it can be tuned for some other architecture or any other specific device according to need is challenging task. Designing an efficient object detection model require computation resources and bigger neural network (time consuming) to achieve better accuracy. On other hand, designing an object detection model which has computationally constraint and smaller neural network that runs a bit faster, at the cost of a little bit of accuracy. EfficientNet model gives a way where model automatically scale up or down neural networks for a given device. 3.2.Results Images from the ZED camera can be viewed and analyzed using a scripted depth viewer. The visualization utility program has been modified and is now used to export each individual bounding box as its own image. This will be used later for object classification to determine whether a leaf is diseased. The proposed model was tested using five images. The model was successfully identified the leaves in test images. Objective 2. Evaluation of Physiological Changes in Flowering Dogwoods in Drought Conditions in a Container Production System The study was conducted in the outdoor setting of Otis L. Floyd Nursery Research Center, McMinnville, TN during May 2022. Sixty-nine commercially grown dogwoods in container production system were arranged in randomized complete block design. Three representative sample trees from each treatment were chosen to determine the daily water usage before starting of irrigation treatments. Sample trees were fully saturated on the first day and their field capacity (FC) was recorded. Next day those trees were again weighed to determine their evapotranspiration loss (ET) and average difference between FC and ET was calculated for a week to determine the daily water usage. Treatments were considered as control, moderate and severe drought (125%, 25% and 10%, respectively) based on their daily water usage. The trees were covered with saucers to prevent the entrance of rainwater in the container. Mid-day leaf moisture potential (PMS) using Model 600 Pressure Chamber Instrument and Normalized Difference Vegetation Index (NDVI) from Field Scout CM 1000 NDVI Meter were recorded every week for a month. NDVI was also recorded from Sentera single sensor NDVI camera mounted in a DJI Mavic 2 every week to ground truth with the handheld NDVI. At the end of the experiment, fresh root and total plant biomass was recorded. Results: Plant Growth:No significant differences were observed among the treatments for plant height, plant width increase, root biomass and total biomass. Normalized Difference Vegetation Index (NDVI):Significant differences were observed among the treatments for NDVI.NDVI was higher for controls in all 4 weeks among all treatments. In first and second week, moderate drought had similar NDVI as severe drought but higher NDVI in third and fourth week. Mid-day leaf moisture potential (PMS): PMS was statistically similar among the controls in first week. In second, third and fourth week, controls had significantly higher PMS compared to other treatments. Among the drought treatments, moderate drought had significantly higher PMS in second, third and fourth week compared to the severe drought. Objective 3. Assess the economic impact of these technologies on nursery production costs and profitability The survey has been conducted to determine growers' nursery production practices and estimation of the economic impact of those practices.Data will be analyzed and results will be summarized and submitted for peer review.

Publications

  • Type: Journal Articles Status: Published Year Published: 2022 Citation: Neupane, K., Ghimire, B., and Baysal-Gurel, F. 2022. Efficacy and timing of application of fungicides, biofungicides, host-plant defense inducers, and fertilizer to control Phytophthora root rot of flowering dogwoods in simulated flooding condition. Plant Disease. http://dx.doi.org/10.1094/PDIS-02-22-0437-RE
  • Type: Journal Articles Status: Published Year Published: 2022 Citation: Neupane, K., Ojha, V. K., Oliver, J. B., Addesso, K. M., Baysal-Gurel, F., 2022. Integration of control strategies to optimize management of Ambrosia beetles (Coleoptera: Curculionidae, Scolytinae) and Phytophthora root rot (Peronosporales: Peronosporaceae) in flowering dogwoods (Cornalaes: Cornaceae) after simulated flooding. Journal of Economic Entomology. http://dx.doi.org/10.1093/jee/toac093
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Hikkaduwa Epa Liyanage, K., Witcher, A., Khanal, A., and Baysal-Gurel, F. 2022. Assessing the costs involved with dogwood disease management in Tennessee nurseries. 2022 TAS Meeting. Nashville, TN. November 18, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Ojha, V., Oliver, J., Addesso, K., Baysal-Gurel, F. 2022. Comparative efficacy of integrated fungicide, insecticide and blocking agent to manage Phytophthora root rot and Ambrosia beetles in flood stressed flowering dogwoods. 3rd Association of Nepalese Agricultural Professionals of Americas (NAPA) Biennial International Scientific Conference. May 27-29, 2022, Atlanta, GA.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Witcher, A., Baysal-Gurel, F. 2022. Evaluation of Physiological Changes in Flowering Dogwoods in Drought Conditions in a Container Production System. 1890 Research Directors Symposium. April 2-5, 2022. Atlanta, GA.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Witcher, A., and Baysal-Gurel, F., 2022. Early season monitoring of drought induced physiological changes of flowering dogwoods in container production system. The 44th Annual University-Wide Research Virtual Symposium, 2021. March 28-April 1, 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Ojha, V., Oliver, J., Addesso, K., Baysal-Gurel, F. 2022. Integrated management of ambrosia beetles and Phytophthora root rot of flowering dogwoods in a simulated flooding condition. 2022 joint Southeastern branch & APS-CD Meeting. March 26-30, 2022. San Juan, Puerto Rico.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Witcher, A., Baysal-Gurel, F. 2022. Assessment of physiological changes to monitor pests and diseases of container grown flowering dogwoods in drought condition. 10th International IPM Symposium. February 28-March 3, 2022. Denver, CO.
  • Type: Other Status: Other Year Published: 2022 Citation: Baysal-Gurel, F. 2022. Vascular Streak Dieback of Redbud and Other Important Woody Ornamental Disease Updates. TNLA Field day. McMinnville, TN. 25 Aug 2022.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Neupane, K., Witcher, A., Baysal-Gurel, F. 2022. Monitoring of drought induced physiological changes of Cornus florida grown in container production system. 99th Southern division APS hybrid meeting. March 7-10, 2022. Chattanooga, TN.


Progress 01/01/21 to 12/31/21

Outputs
Target Audience:This integrated projectwill result in an UAS to improve nursery production,with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collectfoliar disease, abiotic stressand plant measurement datawithin a nursery production environment. Nursery producers, industry professionals as well as undergraduate students will have access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension and research products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business which will address another NIFA priority area of "Rural communities for the 21st century (expanded opportunities for small businesses and small farms)". Changes/Problems:The major issue was caused by COVID-19 which led to the grounding of our program during a large portion of this reporting period. As the research and outreach efforts impacted, those required us to request extension for the project. What opportunities for training and professional development has the project provided?Graduate student has received drone usage trainings and received his drone pilot license.Krishna Neupane was able to finish imagingscientist internshipwith Monsanto Technology LLC "Bayer" in Hawaii- Kihei for 10 weeks- Summer 2021.Student learnt experimental design, data collection, tabulation and analysis. How have the results been disseminated to communities of interest?The products, results and impacts of this integrated project were disseminated through a number of methods and to a number of target audiences.Presentations. The project team was involved in developing workshops, training sessions/demonstration field days, research and Extension symposium/meeting presentations.Publications. eXtension publications, fact sheets, Extension articles and technical bulletins were developed and the target audiences were nursery producers, Extension agents, students, scientists and other stakeholders. What do you plan to do during the next reporting period to accomplish the goals?Objective 1. Design/build an UAS to operate within a nursery production environment. This research is being conducted using an interactive process approach, which needs continuous improvement of the design, concept, hardware, and software systems. The repeated trials and evaluations help us to get closer to our final prototype. The remaining final iterations and the development are summarized in the next sections. (1) Hardware The gripper design will be finalized and tested on collecting dogwood tree leaves in the nursery. (2) Software The user application will be finalized and tested with the users on the workstation. The database system and the data collection system will be optimized for faster data processing and gathering. The leaf detection system and the disease detection system will be improved with more data from the field. (3) Evaluations Final evaluations will be made and analyzed. The performance of the overall system will be recorded and analyzed in terms of time, speed, correctness of measurements and efficiency. Objective2.Evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements Drought stress experiment and nutrient stress experiment will be conducted in spring 2022 for potential detection using UAV. Objective 3.Assess the economic impact of these technologies on nursery production costs and profitability Survey will be conducted to determine growers' nursery production practices and estimation of the economic impact of those practices and the data will be analyzed. Objective4.Technology and information transfer to practitioners Outreach activities will be conducted.

Impacts
What was accomplished under these goals? Objective 1. Design/build an UAS to operate within a nursery production environment The objective of this project is to design and develop an UAV system which is able to create a 3D structure of a tree and find the leaves on this tree for further disease analysis. The design and development part can be divided into 3 main parts: (1) hardware, (2) software and (3) evaluation. Progress and status of each part will be discussed in the next sections (1) Hardware The main structure of the UAV system has been designed and the appropriate hardware has been selected. The list of the equipment purchase is at the below: ZED Depth Sensor 2K Stereo Camera Mavic 2 Pro Drone NVIDIA Jetson Nano Auxiliary and battery setups GPS kit for Jetson Nano Our ZED Depth Camera and Nvidia Jetson Nano computer system have been attached on the bottom of the UAS and we started to getting data. The GPS module has been added to the prototype and been used to capture the location of the trees. The workstation computer has been ordered and will be used to store the data. The user is going to use a user-friendly application to reach to the database. On the other hand, Mavic 2 Pro is the drone that can carry 2 lbs. of load, consequently we were not able to add the gripper to the existing prototype. Most drones are built for passive applications such as taking pictures or videos instead of active applications like manipulating, collecting leaves or carrying items. The drones' having limited payloads and experiencing difficulties in flying around dynamical environments are the main reasons why the drones have not been tested or used in an active application such as leaf sample collection. We are testing two-finger and three-finger grippers in the unsafe flying mode. Flying in unsafe mode created some issues and the drone crushed many times into the trees. We are searching for different gripper types and mounting techniques to increase the safety of the flight and improve the leaf sample collection. (2) Software We are working on a user-friendly application software which can be used by a user with limited computer experience. The user will be able to do analysis on the data gathered from the UAV. The software will be used to measure the tree dimensions, autonomously find the leaves, detect the diseases on these leaves, show user the previous data and the location information of the trees. We are still working on finalizing this software. (3) Evaluation We continue evaluating the hardware and software system to improve the performance of the UAS. We have already started the design iterations considering the outcomes of field trials and changed the structure of the system and sensors. For example, the new depth camera system, ZED Depth Sensor 2K Stereo Camera needs a faster data bandwidth to load the data and only the small size high performance computing device was the Jetson Nano. We have purchased it and started developing our algorithms with this new system. We are also working on improving the leaf detection and disease detection algorithms. Objective 2.Evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements Two drought stress experiments (spring and fall) were conducted to evaluate the physiological changes induced due to drought condition using dogwood plants grown in containers. In a field setting, trees were organized in a randomized complete block design. Three different irrigation treatments were applied at 125%, 25%, and 10% (control, moderate and severe drought, respectively) of their daily water usage (evapotranspiration). Physiological parameters like Normalized Difference Vegetation Index (NDVI) and leaf moisture potential (PMS) were collected every week for one month. Plant growth data were collected at the beginning and the end of the study. NDVI collected from a handheld NDVI meter and a Sentera NDVI sensor mounted in a UAV were correlated for ground truthing. In the study, there were no significant differences in plant total and root fresh biomass, however, control plants had the greatest height and width. NDVI from the handheld NDVI meter was significantly higher for control as compared to other treatments in 7th, 14th, 21stand 27thday. The correlation between handheld and Sentera NDVI were 91%, 93%, 84% and 88% in respective weeks. No significant differences were obtained for PMS on day 7 but was greatest for controls on 14th, 21stand 27thday. A separate study was initiated in May evaluating the effect of different fertilizers and a micronutrient fertilizer on dogwood growth in containers. Briefly, container-grown (15 gallon) dogwood plants were treated with 4 different controlled release fertilizers with and without the addition of a micronutrient fertilizer. Plant growth (height, width, and stem diameter) and SPAD (leaf greenness) were measured monthly while substrate pH and electrical conductivity were measured every two weeks. NDVI did not detect any differences among treatments, thus these measurements were not continued. Leaf tissue nutrient content was analyzed at the end of the study. Data will be analyzed and results will be summarized and submitted for peer review. Results will also be used to establish a nutrient stress project for potential detection using UAV in spring 2022. Objective 3.Assess the economic impact of these technologies on nursery production costs and profitability IRB approval has been received from TSU and Drs. Baysal-Gurel and Khanal are finalizing the survey questionnaire to determine growers' nursery production practices and estimation of the economic impact of those practices.

Publications

  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Neupane, K. and Baysal-Gurel, F. 2021. Automatic Identification and Monitoring of Plant Diseases Using Unmanned Aerial Vehicles: A Review. Remote Sensing 13, 3841. DOI: 10.3390/rs13193841.
  • Type: Journal Articles Status: Published Year Published: 2021 Citation: Neupane, K., Alexander, L. Baysal-Gurel, F. 2021. Management of Phytophthora cinnamomi using fungicides and host plant defense inducers under drought conditions: A case-study of flowering dogwood. Plant Disease. https://doi.org/10.1094/PDIS-04-21-0789-RE.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Neupane, K., Witcher, A., Baysal-Gurel, F. 2021. Measurement of drought induced physiological changes in flowering dogwoods in container production system. 2021 TAS Meeting. Tennessee Tech University, Cookeville, TN. November 6, 2021. (Oral presentation first place).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Neupane, K. and Baysal-Gurel, F. 2021. Efficacy of preventative fungicides and host plant defense inducers to manage Phytophthora root rot under drought conditions. MANRRS Regional Competition. Masters Division. Virtual. October 12, 2021. (Oral presentation first place).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Neupane, K. and Baysal-Gurel, F. 2021. Management of Phytophthora cinnamomi using fungicides and host plant defense inducers under drought conditions. Annual Meeting of the American Phytopathological Society  Plant Health 2021 online. August 2-6, 2021 (Poster presentation) (complimentary registration award from Bayer ($269)).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Neupane, K., and Baysal-Gurel, F. 2021. Efficacy of fungicides and biofungicides to manage Phytophthora cinnamomi under drought condition. The 43th Annual University-Wide Research Virtual Symposium, 2021. March 22-26, 2021 (Oral presentation).
  • Type: Conference Papers and Presentations Status: Published Year Published: 2021 Citation: Neupane, K., and Baysal-Gurel, F. 2021. Efficacy of fungicides, biofungicides, host plant defense inducers and fertilizer to manage Phytophthora root rot of dogwood under flooding condition. 98th Southern division APS virtual meeting. Feb 15-19, 2021 (Southern division APS meeting scholarship award) (Oral presentation).


Progress 01/01/20 to 12/31/20

Outputs
Target Audience:This integrated projectwill result in an UAS to improve nursery production,with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collectfoliar disease, abiotic stressand plant measurement datawithin a nursery production environment. Nursery producers, industry professionals as well as undergraduate students will have access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension and research products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business which will address another NIFA priority area of "Rural communities for the 21st century (expanded opportunities for small businesses and small farms)". Changes/Problems:Due to Covid 19, there were delay on the constructions of nursery production sites.Due to Covid 19 restrictions in the main campus, PIs and graduate students had and still have limited access to their Labs. And again, due to Covid 19, there were limitations to conduct outreach/extension efforts in 2020. One of the graduate studentleft the graduate program due to his health conditioninSpring 2020. New graduate student was hired in Fall 2020. Those issues, restrictions and graduate student change may result delay in the project. What opportunities for training and professional development has the project provided?Graduate students have received drone usage trainings and received their drone pilot licences. How have the results been disseminated to communities of interest?Extension Material:One extension article wasdeveloped for nursery producers, industry professionals, potential service providers, Extension educators, scientists, general public and students. Presentations:The project team was involved in developing 2 workshops for the Extension agents and growers, and one master gardener training sessions/demonstration field days. What do you plan to do during the next reporting period to accomplish the goals?Objective 1.Design/build an UAS to operate within a nursery production environment We still continue on evaluating the hardware and software system to improve the performance of the UAS. We have already started the design iterations taking into account the outcomes of field trials and changed the structure of the system and sensors. For example, the new depth camera system, ZED Depth Sensor 2K Stereo Camera needs a faster data bandwidth to load the data and only the small size high performance computing device was the Jetson Nano. We have purchased it and started developing our algorithms with this new system. Still depending on the field experience, we may need to change the design and continue design iterations. In the vision-based DCNN leaf disease detection, we utilize a pre-trained model to build upon. The model we use is SSD ResNet50 V1 FPN 640x640. This model provides a good speed and performance. The model will provide a good starting point for training our network. The first step is to download the pre-trained model from TensorFlow's Model Zoo. When we download this pre-trained model it comes with Checkpoints, Saved Model asset and variables, and then a pipeline configuration file. The pipeline configuration file is then modified to suit our needs. In the code below, we had to change the batch size down to two because of the GPU running out of memory. To ensure that we did not lose training performance, the number of steps were increased to 50,000. The preliminary test images are results that show the "HealthyLeaf" and "DiseasedLeaf" detections, can be seen in Figures 6-8. Figure 6. Test Results Figure 7. Test Results Figure 8. Test Results. Objective 2.Evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements A preliminary drought stress trial was initiated in September to establish irrigation protocols for inducing effective drought response to container-grown dogwood plants. Twenty-four dogwood plants (grown in 5 gal containers) were placed on a gravel pad and each container was fitted with a pressure compensated spray stake (Gray Double Spray; Netafim USA) to supply irrigation. All plants were irrigated daily to saturation for one week and plants were weighed 1 hr following irrigation and 1 hr prior to irrigation the following day to calculate average daily water use/loss. After one week, irrigation treatments were developed by applying 125%, 50%, or 0% of the daily water loss to plants (8 reps/treatment). Plants are being monitored for visual symptoms of drought stress, along with quantitative measurements of leaf chlorophyll using a handheld NDVI meter and a handheld SPAD meter. Additionally, aerial images of tree canopies will be recorded using a UAS fitted with a multispectral camera and NDVI will be analyzed using Sentera Field Scout software. Based on results of the preliminary trial, drought stress irrigation treatments will be finalized and a full-scale drought stress trial will be conducted in spring 2021. Objective 3.Assess the economic impact of these technologies on nursery production costs and profitability Drs. Baysal-Gurel and Khanal will be conducting survey to determine growers' nursery production practices and estimation of the economic impact of those practices. Objective4.Technology and information transfer to practitioners

Impacts
What was accomplished under these goals? Objective 1. Design/build an UAS to operate within a nursery production environment (Second Year) The second-year objective of this project is to design and develop an UAS system which will be able to create a 3D structure of a tree and find the leaves on this tree for further disease analysis. Progress and status of each part will be discussed in the next sections. (1) HardwareMavic 2 Pro is the drone that can carry 2 lbs. of load. Our ZED Depth Camera and Nvidia Jetson Nano computer system have been attached on the bottom of the UAS and we started to get the first data. The algorithms and decisions on methodology have been discuss in the next software section in detail. The design and implementation of the gripper for the drone is in progress. In order to safely, design and do the trials, a new and cheaper drone has been purchased. This drone will be used only for testing the gripper, that will be mounted later at the bottom of the Mavic 2 Pro Drone.The gripper system is a stand-alone system that could be use for any UAS system. This system has its power, controller and the wireless communication system. (2) SoftwareAn architectural design of the Vision based DCNN leaf disease detection systems built to find the diseased leaves. The system architecture design will give an overall view of the system. The detailed design will delve deeper into each component of our system. This will include the labeling software, leaf detection subsystem and the leaf disease detection subsystem. System Architecture A collection of various dogwood leaf images is fed into the LabelImg software to annotate each image to prepare it before inputting it into TensorFlow. Once in TensorFlow it creates a DCNN model to detect dogwood leaves. After this is done, another set of images that contains foliar diseased leaves is then run through LabelImg. Then, the annotated information is run through TensorFlow and get a DCNN model from the data. Finally, the models are to be combined to allow for leaf detection and foliar disease detection on the leaf. Once this is finished an image from the UAS and distinguish the leaf and determine if the leaf is diseased. Labeling LabelImg is used for the purpose of annotating the images. The images under both the train and test sections are used for generating annotations and with the help of LabelImg. The first requirement is the need of ample amount of data to avoid over-fitting. This means ample number of images are needed for each section of detection that is performed in order to avoid getting an overfitted model. This is ensured that about 100 images are available for each of the class of the model that is considered. LabelImg is activated under the environment that is used for annotating the images. Subsequently, LabelImg is provided the directory under which all the images are stored. The annotation of images are carried out. After the process, a bunch of XMLs are generated where each XML is produced for each of the images present under that directory. The next task performed is the division of image dataset for training and testing purposes. The images are divided in the 90% - 10% ratio such that 90% of all the images under each class are used for training the model and the rest for testing the model accuracy. The images are stored under different directory for convenience. The corresponding XMLs are also saved with the respective images under the train and test directories. Then, a Label Map is constructed because TensorFlow requires it to map each of the labels used in the dataset to an integer value. This Label Map gets used for both the training and testing purposes. Consequently, TensorFlow records are also created. This is undergone in a twostep process. The first step is converting the individual XML files generated after annotating the images to a unified CSV file for each dataset. This means generating a unified CSV for both the Leaf Detection data set and another for the Leaf Disease Detection data set. The second step is conversion of these unified CSV files to each dataset to record files (the TFRecord or the TensorFlow record format). This results in two record files for each data set i.e. test record and train record file for the Test and Train data set accordingly. Leaf Detection Subsystem The convolution neural network (CNN) is associated with several neural network layers. However,deep neural network architectures have been recommended for implementation to help in enhancing the accuracy of the object detection. The overall system receives image frames for a tree in order to detect the occurrence of disease. The disease detection system can analyze individual leaves located on the tree. That's why, it is an obvious requirement to identify leaves on the received image frame, which belongs to the interested tree. In order to detect visible leaves on the image frame, an object detector could be constructed determine the locations of the leaves. The object detector utilized in this project is a Single ShotDetector (SSD) [1]. This architecture is quite popular in object detection field, because it supports great running time gains over other popular architectures such as Region-Based Fully Convolutional Networks (R-FCN), Faster-RCNN. ?R-FCN and RCNN executes region detection and region classification steps separately. First, they perform a region detection operation via a detection network in order to obtain possible regions, then process the regions to obtain classification results for each one. However, the advantage of SSD comes from the idea of processing these steps at the same, that means simultaneously acquiring predictions of regions and corresponding classification results. This idea yields an incredible speed over region proposal-based architectures. Disease Detection Subsystem ?Neural networks have seen their application in the field of agriculture. The aspect has influenced the development of efficient independent systems needed to help in object and disease detection in plants. For instance, the deep convolution neural network (DCNN) has been established to assist in detecting plant diseases.However, several issues may be witnessed when utilizing DCNN. For instance, complex field conditions may be challenging to identify and detect using this approach. Remarkably, this is because several infections and pathologies may be existing in the same image or surrounding objects. Objective 2.Evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements A preliminary drought stress trial was initiated in September to establish irrigation protocols for inducing effective drought response to container-grown dogwood plants. Twenty-four dogwood plants were placed on a gravel pad and each container was fitted with a pressure compensated spray staketo supply irrigation. All plants were irrigated daily to saturation for one week and plants were weighed 1 hr following irrigation and 1 hr prior to irrigation the following day to calculate average daily water use/loss. After one week, irrigation treatments were developed by applying 125%, 50%, or 0% of the daily water loss to plants . Plants are being monitored for visual symptoms of drought stress, along with quantitative measurements of leaf chlorophyll using a handheld NDVI meter and a handheld SPAD meter. Additionally, aerial images of tree canopies will be recorded using a UAS fitted with a multispectral camera and NDVI will be analyzed using Sentera Field Scout software. Objective 3.Assess the economic impact of these technologies on nursery production costs and profitability IRB approval has been received from TSU andDrs. Baysal-Gurel and Khanal are working on a survey questionnaire to determine growers' nursery production practices and estimation of the economic impact of those practices.

Publications

  • Type: Other Status: Published Year Published: 2020 Citation: Baysal-Gurel, F. 2020. What Can Drones Do for You? Chase Digest October 2020 Issue Volume 8(10).


Progress 01/01/19 to 12/31/19

Outputs
Target Audience:This integrated projectwill result in an UAS to improve nursery production,with the capability to move above the plant canopy, between rows of plants, around individual plants, and hover directly beside a plant to collectfoliar disease, abiotic stressand plant measurement datawithin a nursery production environment. Nursery producers, industry professionals as well as undergraduate students will have access to hands-on experience as well as knowledge of UAS uses for nursery production. The utilization of the UAS and information delivery through Extension and research products will increase the likelihood of adoption by nursery producers and service providers by providing opportunities for small business which will address another NIFA priority area of "Rural communities for the 21st century (expanded opportunities for small businesses and small farms)". Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided? Nothing Reported How have the results been disseminated to communities of interest?In order to find proper direction and fully understand all possibilities and approaches associated with this project, a comprehensive literature review was conducted. This literature review--titled "Unmanned Aircraft System (UAS) technology and applications in agriculture"--was published in volume 9 of Agronomy and selected as the front cover for issue 10. This review summarizes the current literature involving UAS usage in agricultural applications and provides a foundation for notonly this project, but many other future research projects of similar scope. What do you plan to do during the next reporting period to accomplish the goals?We still continue on evaluating the hardware and software system to improve the performance of the UAS. We have already started the design iterations taking into account the outcomes of field trials and changed the structure of the system and sensors. For example, the new depth camera system, ZED Depth Sensor 2K Stereo Camera needs a faster data bandwidth to load the data and only the small size high performance computing device was the Jetson Nano. We have purchased it and started developing our algorithms with this new system. Still depending on the field experience, we may need to change the design and continue design iterations. We will evaluate the UASfor early detection and identification of foliar diseases, determination of abiotic stress factors, and plant growth measurements in outdoor settings.

Impacts
What was accomplished under these goals? Objective 1.The first-year objective of this project is to design and develop an UAS system which will be able to create a 3D structure of a tree and find the leaves on this tree for further disease analysis. The design and development part can be divided into 3 main parts; (1) hardware, (2) software and (3) evaluation (see Figure 1). (1) Hardware The main structure of the UAS system has been designed and the appropriate hardware has been selected. The list of the equipment purchase is at the below: ZED Depth Sensor 2K Stereo Camera Intel® RealSense™ Tracking Camera T265 Mavic 2 Pro Drone Velodyne Puck LITE 3D Laser Scanner Raspberry Pi 3 NVIDIA Jetson Nano Auxiliary and battery setups Mavic 2 Pro is the drone that can carry 2 lbs of load. Our ZED Depth Camera and Nvidia Jetson Nano computer system have been attached on the bottom of the UAS and we started to get the first data. (2) Software 3D Tree Modeling Photogrammetry Reconstruction Photogrammetry is a computer vision and computer graphics process to measure and interpret the photographic images. Must popular application is the acquiring 3D measurements from 2D image data. This process called reconstruction. Structure-from-Motion (SfM) is an algorithm that acquires 3D reconstructed version of the interested object. The process based on the multiple image motion observed from a camera in motion (Schonberger, 2016). To obtain a 3D reconstructed object from a single moving camera (mostly around the 3D object), it is required to take pictures of the object in various positions and orientations. The number of images depends on the object size and desired quality. The crucial part is that the sequentially taken images must contain common parts. As a common practice, the common visual part in the image plane should be higher than 70%. To conduct the reconstruction, it requires to perform sequential steps as shown in the given block diagram. The input of this sequential process is the recorded images mentioned above. Feature detection, extraction and matching are common practice in computer vision area. There are numerous algorithms to execute these processes. The main idea is to determine the distinguishable points located on the images. In the extraction phase, these points are converted into vector from to represent the image plane characteristics of the points such as intensity, orientation, scalability etc. Popular algorithms are SIFT (Scale-Invariant Feature Transform), SURF (Speeded-Up Robust Features), Akaze (Tareen, 2018). The information stored in the vector from depends on the algorithm that is utilized on detection phase. After obtaining the feature information for all images, the matching procedure is required to be carried in order to determine which features are matched on other images. The most accepted matching algorithm is FLANN (Fast Library for Approximate Nearest Neighbors) (Alhwarin, 2010). Next phase of the photogrammetry process is the bundle adjustment, which utilized the matched features acquired in the previous phase. It is defined as optimizing 3D coordinates representing the visual characteristics, by estimating the motion parameters of the perception device (camera) through 3D cartesian space. Through the bundle adjustment process, the relation between the images framesare estimated (Wu, 2011). This estimation is a numerical optimization process and it is the most important part of all reconstruction process. At the end of the bundle adjustment, a sparse point cloud is generated for the feature's points detected. By utilizing the estimated camera location, relations in between the images, the features could be densified by increasing the matching point number for only the sequential image groups. Increasing the image plane points and corresponding matches yields a densified point cloud via triangulation. Densified point cloud is utilized to generate the mesh structure of the constructed object to represent it more accurately in 3D form as shown in the following figure. The last to steps are applied to colorize (Texture Mapping). When we apply the described operations in order to reconstruct a tree, unfortunately, the acquired 3D model does not represent the tree properly (few pictures shown in following figures). There are major scalability errors occur. The main reason of that situation is that the feature extraction & matching phases are failed in the progress. The features detected in taken images are very similar with the other features (They cannot be distinguishable). In another experiment of the concept, the problem is directly related with the first phase of the process pipeline. The feature detection algorithms can not find proper features on the tree itself. Most proper features are detected on the environment, instead of the tree. This problem fails the overall reconstruction process. RGBD (Red-Green-Blue-Depth) Camera Reconstruction As said, humans can have a three-dimensional perception of the world through the eyes due to the difference observed in the images formed in left and right retinas. In the imaging process, the images sent to the brain from each eye are not the same, with a slight difference in the position of the objects due to the separation between the eyes, which form a triangle with the scene points. Thanks to this difference, by triangulation the brain can determine the distance (depth) that the objects are in relation to the observer position. The implementation of stereo vision in most RGBD cameras uses this basic principle to recreate a 3D scene representation based on the two images of it taken from different viewing points. This is known as stereo reconstruction. The principles of the reconstruction process by using an RGBD camera is very similar with the regular single camera. However, additional depth information, which is estimated via stereo vision, allows the process to generate dense point cloud for the instantly seen frame pairs directly. This property, eliminates the requirements of strong features which are utilized to estimate the camera location for each frame . Main concern is to combinethe sequentially taken point clouds within the different time instances in order to generate an overall 3D point cloud for the object that is going to be reconstructed. Iterative Closest Points (ICP) algorithm is the solution of the issue. After the registration, mesh generation and texture mapping is applied to obtain the final model. Objective 2. (1) Literature review In order to find proper direction and fully understand all possibilities and approaches associated with this project, a comprehensive literature review was conducted. This literature review--titled "Unmanned Aircraft System (UAS) technology and applications in agriculture"--was published inAgronomy and selected as the front cover for issue 10. (2) Creation of image databases A promising avenue for disease detection relies on utilizing machine learning techniques such as convolutional neural networks to classify healthy and diseased tissue. In order to train these algorithms, there must be a significantly large database of images (preferably >1000 images). Unfortunately, there were not any large databases of flowering dogwoodleaves available online; therefore, we created our own database for this project. As of now, we have well over a thousand images taken from dogwoods at our station, as well as surrounding nurseries. We have also built a smaller database (~150 images) of aerial shots taken of containerized dogwoods from various backgrounds. These images were annotated with XML anchor boxes locating the tree canopies in order for our algorithms to learn to detect individual trees.?

Publications

  • Type: Journal Articles Status: Published Year Published: 2019 Citation: Hassler, S., and Baysal-Gurel, F. 2019. Unmanned aircraft system (UAS) technology and applications in agriculture. Agronomy. Agronomy 2019, Vol: 9, 618. DOI: 10.3390/agronomy9100618.