Progress 09/01/23 to 08/31/24
Outputs
Target Audience:Golf course superintendents, turfgrass sod producers, landscape managers, turfgrass scientists, turfgrass breeders. Changes/Problems:The linear gradient irrigation trial at the University of California, Riverside faced several challenges, including delays in starting data collection due to difficulties in recruiting graduate students for the project. For the UAS-remote sensing and ground measurements of the bermudagrass population at the University of California, Riverside, the major problem has been associated with the frequency of data measurements (both UAS- and ground-based). A lack of personnel resulted in ground measurements being performed once a month, while UAS-based measurements were performed once per week. Moving forward, we will aim to collect ground data on a weekly-basis.Despite these setbacks, we believe the delays will not require major changes to our approaches or the completion of our proposed objectives. What opportunities for training and professional development has the project provided?The project has provided training opportunities for students including aspects of experimental design, irrigation design, and turfgrass maintenance associated with the planning and construction of the linear gradient irrigation field required for the project. The project provided the opportunities for training graduate students and undergraduate students in remote-sensing technology with drone carrying multispectral, RGB, and hyperspectral sensors for high throughput phenotyping of turfgrass and evaluation of plant responses to deficit irrigation, as well as instruments for ground measurements of turf performance, canopy temperature, and soil water content. The project has provided an opportunity to train a postdoctoral research associate (Ved Parkash), and well as graduate students (Kumar Vaibhav and Manveer Singh) to utilize remote sensing platforms for data collection and implementation in high-throughput plant phenotyping. It also trained graduate students on survey design, IRB, and think-aloud techniques. The project provided engineering and computer sciences design opportunities for high-school student (one student), undergraduate (four students), and graduate (two doctoral and one master) students. How have the results been disseminated to communities of interest?Research information and data were disseminated through academic conference, extension and education activities. We have presented updates of our research at Field Day events, golf course superintendent meetings, sports field manager association meetings, and various other research conferences at state and national levels. This included the Georgia Golf Environmental Foundation summit (January 17th, 2024), the UGA-Griffin Turfgrass Research Field Day (August 7th, 2024). We made presentation and demonstration of the research projects to over 800 turfgrass practitioners and professionals at Rutgers Turfgrass Research Field Days on July 30 and 31, 2024 and Dec. 6-7, 2023 at the New Jersey Green Expo. Presentations were also given to turfgrass sod growers at meetings of Cultivated Sod Association of New Jersey on February 5, 2024 and July 31, 2024. The mobile ground robot was demonstrated at the Rutgers Turfgrass Days on July 30-31, 2024. The robotic systems design and testing were also presented at a few institutions by co-PI Jingang Yi. The presentation by William Errickson, Spencer Gaylert, Yanhong Jin, Cara Cuite, and Bingru Huang (2023) "Needs Assessment for Remote Sensing- and Machine Learning-guided Precision Turfgrass Irrigation Programs: Findings from a Socioeconomic Survey." Received the Outstanding Poster Award by the Crop Science Society at the 2023 ASA-CSSA-SSA International Annual Meeting in St. Louis, MO, October 29-November 1st, 2023. What do you plan to do during the next reporting period to accomplish the goals?Objective 1: Rutgers University: We will process remote-sensingimages and analyze ground data required by Siemens company to develop the DDS for cool-season turfgrasses. The data will also be analyzed to write referereed journal publications. We will conduct the remote sensing high thoughput phenotyping for a larger population of cool-season turfgrasses to validate the DSS developed based on the test set oftall fescue. The robot will be fully deploiyed and take measurements of vegetation indices with RGB caemra and canopy temperature with an infrared thermal camera. University of Georgia - During the upcoming reporting period, we will continue collecting validation data sets on the F1 zoysiagrass field breeding trials in Tifton, GA. These data sets will include RGB and multispectral remote sensing data as well as environmental weather data. In Griffin, additional data will be collected on the LGIS field to refine models. Warm-season turfgrass trials at University of California-Riverside - At the UCR turfgrass trial, irrigation gradients will be applied to study bermudagrass under different levels of water stress induced by the Linear Gradient Irrigation System. This study will involve collecting both remote sensing data from Unmanned Aircraft Systems (UAS) and ground-based measurements of plant and soil factors to train a Decision Support System (DDS) for precision irrigation management in turf. Data collected from the UAS remote sensing and ground measurements for the bermudagrass population will be submitted to Siemens Technology to further train and validate machine-learning models for predicting turf quality and selecting drought-tolerant lines in bermudagrass and other warm-season turfgrasses. Cool-season turfgrass trials at Rutgers University - In the coming year, we will focus on analysis of image and data collected in 2024 and working with Siemens's team to develop and test the machine-learning models for phenotyping and irrigation management. Robotic phenotyping at Rutgers University - We plan to conduct extensive field imaging survey by using the developed ground robot. The integration of the ground and aerial robotic systems will be also planned in the next year to improve the data quality and machine learning-based DSS performance. Objective 2: We will complete the full pipe-line of the application that will have a docker image with the application packed in for easier deployment of the Decision Support System (DSS), user do not need to have the knowledge of the development libraries such as Python/Pytorch. Also, we will update the system to return a binary decision for input RGB, Hyper/Multi-spectrum data so that user can determine the binary decision threshold. We will created documentation about the DSS on how to use the system, how to train and deploy the model and more. We will perform on-site tests on golf courses, breeding trials, or sod farms to validate the DSS for high throughput phenotyping and precision irrigation management program. Field sites are currently being identified for the validation of precision irrigation models and quantification of water saving potential. These trials are expected to begin in the spring of 2025.We will start site assessement and starttesting the DSS for precision irrigation management on golf courses and sod farms. Objective 3: a) Continue to conduct analyses of the first round of survey data; b) Continue literature review on PIM and water reservation technology adoption We conducted choice experiments for this USDA-funded research project, aiming to quantify the importance of different factors in adopting a novel mobile remote sensing-based precision turfgrass management technology. The team is planning to analyze the data from these choice experiments. c) Work on manuscripts based on the survey data. d) Disseminate research findings to communities of interests
Impacts
What was accomplished under these goals?
Objective 1a-b. Rutgers University: Remote-sesin images collected in 2023 were processed and submitted to Siemens company to develop the machine-learning models, including images from remote-sesing of irrigation experiment for creeping bentgrass and kentucky bleugrass collected iin the Lienar Gradient Irrigation System (LGIS) duringJune-July and high thoughput phenotyping of drought tolerance for42 tall fescue varieties exposed to drought stress in a rainout shelter duringAugust-October, 2023. In20024, deficit irrigation experiment was performedfrom June 11th to June 28th.Ground measurements and remote-sensing image analysis for vegetation indices were performed on creeping bentgrass and Kentucky bluegrass in field plots irrigated with the Linear Gradient Irrigation System (LGIS), which was activated once a week to generate soil water gradient across the field site.For creeping bentgrass, NDVI, SIPI, and PSRI showed the highest correlations with turf quality (TQ) among various indices from remote sensing.For SIPI, NDVI, and PSRI, the threshold values for the minimal acceptable TQ level (6) are 0.79, 0.74, and 0.11, respectively, corresponding to 25% soil moisture.For Kentucky bluegrass, NDVI, NDRE, SIPI, and PSRI all showed strong correlations with TQ . The threshold values for the minimal acceptable TQ level (6) are NDVI 0.68, SIPI 0.79, PSRI 0.15, and NDRE 0.33, corresponding to a 23.5% soil moisture level.SIPI in both grass species demonstrated the best consistency across different days of data collection and the strongest overall correlation with TQ, suggesting that SIPI is the top indicator for monitoring turf performance or health status in cool-season turfgrasses. University of Georgia: A soil moisture sensor array consisting of 24 in situ sensors was installed in the Linear Gradient Irrigation System (LGIS) field prior to the growing season in early spring in Griffin.Deficit irrigation trials were conducted in June 2024, with data collected on the linear gradient fields to develop irrigation models. Data collection included RGB, hyperspectral, and thermal-imaging, soil moisture, weather data, visual quality, and leaf water content. During a three-week period from June 7th to June 28th, 2024, soil moisture in the driest parts of the field fell from ~25% VWC down to ~5% VWC. Corresponding leaf water contents ranged from 95% leaf water content down to 67% leaf water content. Similarly, Normalized Difference Vegetation Index values from reflectance data ranged from 0.83 in plots with adequate water down to 0.36 in the driest plots. Remote sensing data, along with environment and ground truthing data will be processed by collaborators to develop models to support the development of a decision support system. Training data sets comprised of RGB and multispectral remote sensing data, along with environmental data, were collected from bermudagrass and zoysiagrass field breeding trials at Tifton, GA and shared with data scientist at Siemens to develop a machine learning pipeline. Bermudagrass trialsincluded 233 genotypes and zoysiagrass trialsincluded 45 genotypes with 3 replications and 220 genotypes with 2 replications.Validation data sets are being collected from an F1 zoysiagrass population, which has shown phenotypic variation in drought stress response through prior work. The population consists of 228 F1 progeny derived from a cross between Zoysia japonica ('Meyer', 4x) and Z. matrella (PI 231146, 4x). The field trial includes the 228 F1 progeny and the two parents, along with 8 commercial standard cultivars. The trial was established at Tifton, GA in 2020 in randomized complete block design with 4 replications. Data from this trial will be shared with scientists at Siemens to validate machine learning models developed previously. ?University of California, Riverside: The Linear Gradient Irrigation System (LGIS) is being evaluated on bermudagrass sod. During the experiment, only the central irrigation line is active, supplying water to the experimental plots. Unmanned Aerial System based remote sensing platform with one drone carrying a multispectral and RGB camera was used to capture images over the course of the study. In a separate study, a bermudagrass population consisting of inter- and intraspecific hybrids was evaluated under drought. Full drought (0% evapotranspiration replacement) was imposed on the population in mid-June and will be continued until October. Robot design and development: Rutgers team developed a new mobile ground robot that includes the robot, navigation and imaging system for performance improvement. The main improvements include: (1) lightweight platform with improved rigidity and less turfgrass compaction due to tire track; (2) improved navigation systems for high-precision localization; and (3) improved imaging systems for the RGBD and thermal cameras. The new robot was tested and demonstrated in Rutgers Turfgrass Day on July 30, 2024. ?Objective 1c. Extended the machine-learning based pipeline to be able to process the newer datasets that includes multi-spectrum data. This updated pipeline has pre-processing module to enhances the data quality for better performance. Also, the model's hyperparameters are updated for the newer datasets. This updated pipeline contains a newly created domain adaption neural network for the handling multi-spectrum data. An AI based annotation tool (Label Studio) was integrated to DSS module for data labeling, image classification, object detection, and segmentation. Updated the visualization interface based on the feedback that integrates test module of the machine learning pipeline and allows the user to access the performance of the machine learning pipeline and to investigate the anomalies in the dataset. Extended backend RESTAPI to support fetching DSS result and visualize the same in UI. Developed a data driven dashboard to visualize various charts and data generated from the DSS. This dashboard will allow the user to drill-down to various levels of data. Updated frontend UI components to visualize the time-series data and also DSS results as charts and graphs Extended PostgreSQL database schema to store weather time-series data, image meta-data, and also the results that can be visualized and generated as reports. Objective 2.Preliminary site assessment was conducted on two golf coursesin NJ for variability in turf performance and soil water content on fairways . Soil water contact exhibited a wide range of variation across each fairway, ranging from 20% to 40% using the TDR meter when the entire fairway has been irrigated for the same amount and duration and the analysis of multispectral images captured using a drone demonstrated large variability in vegetation index, NDVI in two fairways on a public course. The variations in turf performance were correlated to soil water content. This preliminary assessment suggests a great potential for the use of precision irrigation management programs on public golf courses where soil conditions are variable, and irrigation practices or systems may not beefficient. Objective 3.Completed the online survey with IRB and perromfed the online survey with turfgrass professionals; Preliminary summary statistics are conducted based on a total of 146 respondents by August 14th, 2024. A total of 147 respondents completed the online survey and the majority were golf course manager (N = 68; 46.90%). Among the 75 respondents who answered the question regarding their likelihood of purchasing PIM services and devices, 32% indicated that they would purchase annually renewable PIM service, more than one third (37.33%) were inclined to purchase own PIM service, and 30.67% reported that they would not purchase PIM services or devices. These findings suggest significant potential for increasing the adoption of PIM through education and awareness programs, ultimately improve water use efficiency.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Zhang, J., J Maleski, S Khanal, A Webb, BM Schwartz, SR Milla-Lewis, and. B. Huang. 2023. Assessing Drought Responses of a Zoysiagrass Mapping Population Using UAS-Based Hyperspectral Imaging. ASA, CSSA, SSSA International Annual Meeting
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Errickson, W., S. Gaylert, Y. Jin, C. Cuite, and B. Huang. 2023. Socioeconomic Survey of Needs for Remote Sensing-and Machine Learning-Guided Precision Turfgrass Irrigation Programs. Poster presentation. ASA, CSSA, SSSA International Annual Meeting.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Zhang, J. Insight Beyond Eyesight: Application of different remote sensing techniques in turfgrass management. 2024 Lake Wheeler Turfgrass Field Day. August 14. 2024. Raleigh, NC.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Jespersen, D. et al. 2024. Improving recommendations for precision irrigation management. Turfgrass Research Field Day Guidebook, UGA-Griffin
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Yang, H., T, Zheng, J. Yi and B. Huang (2024). Remote sensing and artificial intelligence-guided precision irrigation management program. Presentation. Proceedings of Rutgers Turfgrass Research Field Days ÿ¿ÿ¢ÿ¿ÿ¿ÿ¿ÿ¿ Golf & Fine Turf. July 30, 2024. Hort Farm II, North Brunswick, NJ.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2024
Citation:
Yang, H., S. Bonos and B. Huang (2024). Remote sensing imagery of high throughput phenotyping to select elite turfgrass cultivars. Presentation. Proceedings of Rutgers Turfgrass Research Field Days ÿ¿ÿ¢ÿ¿ÿ¿ÿ¿ÿ¿ Lawn, Landscape and Sports Turf. July 31, 2024. Rutgers Adelphia Farm, NJ.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Yang, H. 2023. Remote sensing for high throughput phenotyping and detection of stress from deficit irrigation for creeping bentgrass. ASA, CSSA, SSSA International Annual Meeting. 2023, p. 149620.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Zhang, J., Maleski, J., and B. Schwartz. 2023. How do drones, sensors, and data science help turfgrass breeders and turf managers? 75th Southeastern Turfgrass Conference. Athens, GA. October 25, 2023.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Huang, B. 2023. Precision Irrigation Management for Sod Production Using Remote Sensing Technology. NJ Green Expo. Dec. 2, 2023. Atlantic City, NJ.
- Type:
Conference Papers and Presentations
Status:
Other
Year Published:
2024
Citation:
Huang, B. 2024. Remote Sensing and Machine Learning for Precision Irrigation. Cultivated Sod Association of New Jersey Seminar. Feb. 20, 2024. Rutgers EcoComplex, NJ.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Maleski, J., Zhang, J., Vines, P. L., & Schwartz, B. M. 2023 Computer Vision for Turfgrass Quality Assessment [Abstract]. ASA, CSSA, SSSA International Annual Meeting, St. Louis, MO. 152574
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Jing Zhang, Jerome Maleski, Sameer Khanal, Amanda Webb, Brian Schwartz, Susana Milla-Lewis, Aaron Patton, Bingru Huang, David Jespersen. Assessing drought responses of a zoysiagrass mapping population using UAS-based hyperspectral imaging. ASA-CSSA-SSSA International annual meeting. November, St. Louis, MO, 2023.
- Type:
Journal Articles
Status:
Published
Year Published:
2023
Citation:
Katuwal, K., H. Yang. and B. Huang (2023). Evaluation of phenotypic and photosynthetic indices to detect water stress in perennial grass species using hyperspectral, multispectral and chlorophyll fluorescence imaging. Grass Research. 3(1).
|
Progress 09/01/22 to 08/31/23
Outputs
Target Audience:turfgrass practitioners, professionals, and turfgrass scientists/breeders Changes/Problems:For the lienar graident irrigation trial at Griffin campus of University of Georgia, the major problems have been associated with delays in the initiation of data collection associated with the construction of the irrigation system, supply chain issues, and recruiting graduate students for the project. We believe the delays will not result in any major changes in approaches or the completion of proposed objectives. The delay in shipping the soil moisture sensors from the manufacture and water flooding of electronic components for the soil moisture sensors installed underground in the linear gradient irrigation field at Rutgers caused postponement of the deficit irrigation for about 3 weeks at Rutgers University. Frenquent and heavy rain in Julne and July in NJ also interupted the deficit irrigation treatments. We have been challenged in terms of getting survey respondents. Less than expected number of participants for the survey responses have been observed; therefore, additionalstates will be targeted as well as other events within the state of GA, NJ, and CA.We are actively working on solutions for this in partnership with the project team. What opportunities for training and professional development has the project provided?The project has provided several training opportunities for students including aspects of experimental design, irrigation design, and turfgrass maintenance associated with the planning and construction of the linear gradient irrigation field required for the project. The project provided the opportunities for training turfgrass producers on the application of remote sensing platform on turfgrass production and for plant breeder students on high-throughput phenotyping and computer vision in plant breeding. Training of graduate students and undergraduate students in remote-sensing technology with drone carrying multispectral, RGB, and hyperspectral sensors for high throughput phenotyping of turfgrass and evaluation of plant responses to deficit irrigation, as well as instruments for ground measurements of turf performance, canopy temperature, and soil water content. Trained graduate students on survey design, IRB, and think-aloud techniques. The project provided design and fabrication experience for undergraduate students. It also trained the graduate students to conduct integrated STEM and plant biology research. How have the results been disseminated to communities of interest?Drought tolerance, the value of precision irrigation management, and research objectives related to this grant were communicated in extension activities. Results relating to remote sensing platform and preliminary results on using computer vision in turfgrass were disseminated through academic conference, extension and education activities. We made presentation and demonstration of the research projects to approximately 400 turfgrass practitioners and professionals at Rutgers Turfgrass Research Field Days on July 25-26, 2023. Yang, H., T, Zheng, J. Yi and B. Huang (2023). Remote sensing and artificial intelligence-guided precision irrigation management program. Presentation. Proceedings of Rutgers Turfgrass Research Field Days - Golf & Fine Turf. July 25, 2023. Hort Farm II, North Brunswick, NJ. Yang, H., S. Bonos and B. Huang (2023). Remote sensing imagery of high throughput phenotyping to select elite turfgrass cultivars. Presentation. Proceedings of Rutgers Turfgrass Research Field Days - Lawn, Landscape and Sports Turf. July 26, 2023. Rutgers Adelphia Farm, NJ. What do you plan to do during the next reporting period to accomplish the goals?Objective 1: Cool-seaon turfgrass trials atRutgers University: Remote-sensing images and ground data collected from the linear gradient irrigation system with creeping bentgrass and Kentucky bluegrass and the drought trials of evaluating drought tolerance of tall fescue cultivars will be submitted to Siemens Technology. The images and data will be used to develop machine-learning models and decision support system (DDS) for precision irrigation programs and high throughput phenotyping of selection of drought-tolerant germplasm. Warm-season turfgrass tirals at University of Georgia: At the UGA Griffin campus, irrigation gradients will be applied looking at bermudagrass under various levels of water stress induced by the linear gradient irrigation system, taking both remote sensing data was well as ground-based measurements of plant and soil factors to train the DDS for precision irrigation management in turf.Tifton Campus will continue to work with the data scientist at Simens to develop DDS for high-throughput phenotyping. For the developement of decision suppoty system: the following activites will be perormed: 1) We will extend the capability of our machine learning pipeline. The new machine learning pipeline will be able to deal with more data modalities such as time series (weather information) and soil moisture.2) We will also facilitate the semi-supervised learning which mitigates the scarcity of data availability. The semi-supervised learning technique attempts to optimize the extraction of latent features from the input data and further improve pipeline output. 3)We will also extend the machine learning pipeline to the full decision support system, which includes the interfaces that receive data from UAV, moisture sensor and weather stations and send control signal to irrigation systems.4)We will integrate the machine learning and data annotation module. and 5)We will prepare the application test environment for everyone to test and provide feedback. Objective 2:Identify test sites and perform the test of the DSS for high throughput phenotyping. The discussion at the annual project meeting preliminarily decided to perform field tests in National Turfgrass Evaluation Programs (NTEP) trials at Rutgers, UGA, and UC-Riverside to develop the DSS for high throughput phenotyping of warm-season and cool-season turfgrass species. Objective 3:1) Continue literature review on cost savings from PIM and other related finding.2)Continue fielding the first-round online survey with turfgrass professionals. 3)Conduct analyses of the first round of survey data. The survey will be made available to attendees at the 75th Southeastern Turfgrass Conference held in Athens, GA and to turfgrass professionals in New Jersey and California.
Impacts
What was accomplished under these goals?
Objective 1a-b. Developing mobile sensing systems to assess soil and environmental conditions and plant phenotyping under different irrigation regimes for both warm-season and cool-season turfgrasses. Establish threshold values of plant-water stress indices to be used for the development of DSS using the mobile sensing systems developed in Objective 1a. Objective 1a: The mobile ground robot system, inclduing the robot, navigation,imaging systems were designed and fabricated. The robot carrying a RGB camear and thermal camera are tested in turfgrass fields. The system is functional to capture digital and thermal images simultanoursly to evaluate plant vegetation indcies and canopy temperature. The drone remote systems have been developed and used in data collection at University of Georgia and Rutgers University. Objective 1b:Cool-season turfgrass trials at Rutgers University - Unmanned Aerial System based platform with multiple sensors with one drone carrying the hyperspectral camera and one drone carrying a multispectral and RGB camera were tested at Rutgers University. The multispectral/RGB system was used for high throughput phenotyping of two cool-season turfgrass for drought tolerance and for the evaluation to two cool-season turfgrass response to water deficit. The water deficit levels with soil water content measured using a time domain reflectometer ranged from 15% (very dry) to 32% (well-watered, soil field capacity) in the study site irrigated with the linear gradient irrigation system, although frequent heavy rainfall in June-July in NJ reduced the soil water content gradient along the irrigation line. UAV-remote sensing and ground measurements were performed from mid-June and will be continued until October. The preliminary analysis of remote-sensing images and ground data collected from June to August demonstrate that selected vegetation indices (i.e. PSRI, NDVI, and SIPI) from drone multispectral imaging were correlated to ground measurements of canopy temperature by hand-held thermal camera and the percent canopy cover and dark green index measured by hand-held RGB camera. Sensitivity of stress detection and correlation to ground data varied among the selected vegetation indices from the drone multispectral imaging, with PSRI being most sensitive to detect water stress earlier (when soil water content declined only to 25% from the full field capacity of 30%) during soil dry down, followed by NDVI and SIPI. Linear gradient Irrigation treatments and remote-sensing and ground measurements are ongoing and will continue until October. Multispectral and RGB carried by the drone were also used to evaluate turf performance for tall fescue exposed to drought stress in a rainout shelter to determine the vegetation indices that are useful for high throughput phenotyping of drought-tolerant germplasm. Ground measurements, including visual rating of turf quality, percent ground cover and dark green index using a hand-held RGB were also performed. Drought stress was imposed in mid-July. The treatments and measurements are ongoing and remote-sensing and ground images and data are being analyzed. Warm-season turfgrass trials at University of Georgia - The linear gradient irrigation system (LGIS) was installed and tested, which can provide the water distribution gradient across the field site and bermudagrass sod was planted and established on this site atthe Griffin campus. Soil moisutre sensors were purchased and being tested for use in the LGIS) trial.Unmanned Aircraft Systems based remote sensing platform with hyperspectral camera systems were used for imaging collection of a population of warm-seaosn turfgrasses.Relevant workflow to process the hyperspectral images and data was developed. Prelimiary results demonstrated that the hyperspectral imaging system was able to effectively predict turf quality for different turfgrass species varying in vegetation charcateristics and drought tolerance. The mainchine-leanrning models using the hypersepctral data developed from one wamr-season turfgrass, such as zoysiagrass may also be usable for another turfgrass speccies, such as seashpare paspalum. Objective 1c. A machine-learning based pipeline that facilitates automatic phenotyping is being developed. This pipeline includes the training module that consumes multi-spectral data, RGB data and phenotyping ratings as well as the test module that predicts phenotyping ratings based on multi-spectral and RGB data.An annotation interface was designed that allows the users to annotate the phenotyping ratings according to RGB images. A visualization interface was deisgend that demonstrates the output from the test module of the machine learning pipeline and allows the user to access the performance of the machine learning pipeline and to investigate the anomalies in the dataset.Web application frame is developed using three systes: Python Flask - backend processing and integration with other core components; Backend interfaces for user management, data upload and data visualization are completed,ReactJS - frontend web interface development and dashboard development. Basic UI functions such as User Management, Data upload, and Basic data visualization are completed; PostgreSQL - database to store data. All the necessary database schema development is completed. Objective 3: Socioeconomic analysis and extension service: Submitted amendments to the IRB application at Rutgers. Worked with UGA and UC to coordinate with their IRBs. Completed additional pilot studies with participants from the turfgrass industry Edited and refined the survey instrument in Qualtrics. Created and refined multiple types of outreach materials to recruit survey participants Have recruited from multiple audiences, including listservs, conference, and workshop attendees. In process of fielding the first-round online survey with turfgrass professionals Literature review on cost savings from PIM and other related findings Preliminary statistics of the survey data:The survey has been completed by 96 turf professionals, including 29 (30.21%) from New Jersey, 27 (28.31%) from Georgia, 5 (5.21%) from Massachusetts, 3 (3.13%) from California.Among the participants, a majority employ underground sprinkler (48.54%) and hand-held hoses (27.18%) for turf irrigation. Notably, they consistently monitor water usage within their turf areas (84.49%). When it comes to guiding their irrigation decisions, the most prevalent methods include using hand-hold soil moisture sensor (18.41%) and relying on personal observation (15.48%). A significant portion of respondents (64.29%) do not currently utilize any precision irrigation management systems (PIM). For those who do employ irrigation sensing methods, 49.23% use manual sensors, while 26.15% have adopted automated sensor systems. This justifies the significance of this study.
Publications
- Type:
Journal Articles
Status:
Accepted
Year Published:
2023
Citation:
Katuwal, K., H. Yang. and B. Huang (2023). Evaluation of phenotypic and photosynthetic indices to detect water stress in perennial grass species using hyperspectral, multispectral and chlorophyll fluorescence imaging. Grass Research (in press)
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
Yang, H., T, Zheng, J. Yi and B. Huang (2023). Remote sensing and artificial intelligence-guided precision irrigation management program. Presentation. Proceedings of Rutgers Turfgrass Research Field Days â¿¿ Golf & Fine Turf. July 25, 2023. Hort Farm II, North Brunswick, NJ.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Yang, H., S. Bonos and B. Huang (2023). Remote sensing imagery of high throughput phenotyping to select elite turfgrass cultivars. Presentation. Proceedings of Rutgers Turfgrass Research Field Days â¿¿ Lawn, Landscape and Sports Turf. July 26, 2023. Rutgers Adelphia Farm, NJ.
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2023
Citation:
T. Zheng, F. Han, and J. Yi (2023). Design of a lightweight autonomous robotic phenotyping system for turfgrass. In Proceedings of 2023 Modeling, Estimation and Control Conference, Lake Tahoe, NV.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Zhang J., Maleski J. and B. Schwartz. 2023. Drone in sod production. Turfgrass producer international education conference. February 21, Nashville TN.
- Type:
Conference Papers and Presentations
Status:
Accepted
Year Published:
2023
Citation:
Jing Zhang, Jerome Maleski, Phillip Vines, and Bingru Huang. Leveraging UAS-based RGB, hyperspectral images and machine learning in turfgrass breeding. AI in Agriculture: innovation and discovery to equitably meet producer needs and perceptions. April 17-19, Orlando, FL, 2023.
- Type:
Other
Status:
Accepted
Year Published:
2023
Citation:
Jing Zhang and Jerome Malesk. 2023. Image-based high-throughput phenotyping and computer vision in plant breeding. Guest lecture in course PBGG 6000 â¿¿Plant Breeding Practicumâ¿¿ at University of Georgia, Tifton.
|
Progress 09/01/21 to 08/31/22
Outputs
Target Audience:Turfgrass professionals, turfgrass breeders Changes/Problems:Delays in the acquisition of UAS remote sensing system (sensors and drone) and soil moisture sensors due to supply shortage and slow shipping process. Delays in field construction due to labor and supply issues have shifted the completion of the linear gradient irrigation system in Griffin, GA. What opportunities for training and professional development has the project provided?Graduate students and a post-doctoral associate were trained in design and using remote sensing and robotic automation technology for high-throughput phenotyping and detection and monitoring of turfgrass responses to drought stress. The project provided design project opportunities to mechanical engineering students. Four engineering undergraduate students have been trained.. Trained garduate student on survey design and IRB. How have the results been disseminated to communities of interest?Available data and information were presented at the international annual meeting of ASA, CSSA, and SSSA. Drought tolerance, the value of precision irrigation management, and research objectives related to this grant were discussed at the 2022 Turfgrass Field Day at UGA Griffin Campus and Rutgers University. What do you plan to do during the next reporting period to accomplish the goals? We will finalize the fabrication of the autonomous mobile robot for the project. We will also test the robot in the lab and also in the field on Rutgers campus. We expect to deliver the robotic systems to the plant biologists to conduct the field testing and usage in the end of the second year. We will conduct experimentsof detecting and monitoring drought responses for bermudagrass at UGA, Kentucky bluegrass and creeping bentgrass at Rutgers University under various levels of water stress in the linear gradient irrigation site. Both remote sensing dataand groundmeasurements of plantand soil factors, as well as weather data, will be collectedto provide the training datasets to Seimens for the development ofthe decision support system for precision irrigation management. High-throughput phenotyping with UAS-based hyperspectral sensing for bermudagrass at Tifton Campus in UGA will continue.to provide training datasets forSiemens to develop the decision support systems for breeding phenotyping. The PDs from Siemenswill further utilize the collected data to train our SADA-GAN model. We will also implement and deploy the designed interfaces, including visual and annotation interfaces which handle the communication and interaction between human and system and machine-to-machine interfaces such as the data access, convey, and storage ones. Conduct the pilot study in November 2022 to pre-test the survey. Complete IRB final review.. Conduct the first-round online survey with turfgrass professionals. The project team will meet monthly to discuss the project. The project team members will meet with the advisory panel members to provide updates and seek advice. An onsite meeting of all project members and the advisory panel members will take place at Rutgers Turfgrass Field Day or UGA Turfgrass Field Day in the summer 2023. A website dedicated to the promotion and publicizing of the project will be established.
Impacts
What was accomplished under these goals?
Designed and fabricated the autonomous mobile ground robotto take imaging signals from various onboard cameras (e.g., regular RGB, thermal and hyperspectral cameras). The design and initial phases of construction of the linear gradient irrigation system (procurement of materials, layout, and grading) were performed at the Griffin site in the University of Georgia (UGA). The recruitment of a graduate student to work on the project was also accomplished. Completed the installation and construction of the linear gradient irrigation system (LGI) and established two cool-season turfgrass species, Kentucky bluegrass and creeping bentgrass on the LGI site. Unmanned Aerial System based platform and hyperspectral camera systems were purchased and tested at Tifton location of UGA and Rutgers University. A thermal imaging sensor and RGB camera were purchased at Rutgers Unviersity. Initial data collection is taking place in the year of 2022 at breeding plots at UGA. A drought stress trial of high-throughput phenotyping was performed in the rainout shelter at Rutgers University with over 100 different tall fescue and perennial ryegrass cultivars and collected data using multispectral and RGB cameras. Conducted a controlled-environment study with bermudagrass and Kentucky bluegrass at Rutgers University to identify phenotypic and physiological indices from hyperspectral, multispectral, and thermal sensors. Collected water distribution data in the LGI system at Rutgers to test irrigation gradients. Implemented a preliminary machine learning model for the decision support system (DSS). We designed the main framework to facilitate communication between the data collector and machine learning core, as well as the database. Designed the concepts of human-machine interfaces such as visualization, annotation, and uploading interfaces. Completed the draft survey and solicited feedback from within the team and from the advisory board. Submitted the IRB application at Rutgers and completed the initial review. Extension specialists from Rutgers, UGA, and University of California - Riverside coordinated the application of IRBs. Compiled the survey in Qualtrics. Designed the pilot study to pre-test the survey. The project team met monthly to discuss the project. The project team members had two meetings with the advisory panel members to provide updates and seek advice.
Publications
- Type:
Conference Papers and Presentations
Status:
Published
Year Published:
2022
Citation:
Katuwal, K., H. Yang, P. Vines and B. Huang (2022). Detection and Monitoring Drought Stress in Kentucky Bluegrass Using Hyperspectral Sensing and Metabolic Biomarkers. Abstract. ASA, CSSA, SSSA International Annual Meeting. Baltimore, MD.
Yang, H. (2022). UAS-based high-throughput phenotyping in assisting drought-resistant turfgrass cultivar breeding. Abstract. ASA, CSSA, SSSA International Annual Meeting. Baltimore, MD.
|