Source: UNIVERSITY OF CALIFORNIA, DAVIS submitted to NRP
HIGH-THROUGHPUT IN-FIELD PHENOTYPING SYSTEMS TO ACCELERATE BREEDING OF CLIMATE-RESILIENT VEGETABLE CROPS
Sponsoring Institution
National Institute of Food and Agriculture
Project Status
COMPLETE
Funding Source
Reporting Frequency
Annual
Accession No.
1011731
Grant No.
2017-67007-26174
Cumulative Award Amt.
$980,000.00
Proposal No.
2016-07986
Multistate No.
(N/A)
Project Start Date
Feb 1, 2017
Project End Date
Jan 31, 2021
Grant Year
2017
Program Code
[A5171]- Breeding and Phenomics of Food Crops and Animals
Recipient Organization
UNIVERSITY OF CALIFORNIA, DAVIS
410 MRAK HALL
DAVIS,CA 95616-8671
Performing Department
Bio and Ag Engineering
Non Technical Summary
Vegetable production in the US is a $17 billion industry, with ten states accounting for 77 percent of all US vegetable sales for fresh market and processing uses: California, Florida, Washington, Idaho, Arizona, Wisconsin, Oregon, Texas, Michigan and North Carolina (USDA-NASS, 2012 Census of Agriculture). The largest crops by acreage are potato, lettuce, sweet corn, watermelon and tomato. Major vegetable crops in the Solanaceae family include potato, tomato, and pepper; this family is the third most economically important food crop plant taxon (solgenomics.net/about/about_solanaceae.pl). Tomato and pepper contribute to human nutrition and food security in the US. Tomato (Solanum lycopersicum) fruits and products made from fruits contribute significantly to the US dietary intake of lycopene, ascorbic acid (vitamin C), and tocopherols (vitamin E); these antioxidants have been reported as possessing cardio-protective and chemo-preventive activities. Pepper fruits are a rich source of carotenoids (provitamin A) and ascorbic acid, and also supply several B vitamins in the human diet. Tomatoes are consumed fresh or processed (as sauce, paste, diced, juice, etc.) and are an excellent source of nutrients for humans. Peppers (Capsicum annuum) are also consumed fresh or as processed products (as sauces, salsa, diced, roasted, dried and ground). Over the past century, numerous methodologies and technologies have been developed and adapted by plant breeders to expedite cultivar development. Recently, genomics technologies, such as DNA-based markers and genome sequencing, have provided breeders the means to select plants with the desired traits based on their genotype, a process known as marker-assisted selection (MAS). Genotyping has become increasingly high-throughput and cost-effective, with major technical advances in genotyping platforms and data management significantly reducing the costs of employing MAS. Despite rapid advances in genomics, in-field trait phenotyping remains a major bottleneck for crop cultivar breeding and wider implementation of MAS and genomics-facilitated breeding. The manual trait phenotyping process is costly, labor- and time-intensive, and often precludes repeated measurements on the same plants, or plots of plants, over time. Cultivar breeding programs require phenotyping of thousands of plants grown in field locations typical for commercial production of the crop. Multiple traits, some of which are genetically complex (e.g., yield, fruit quality, abiotic stress tolerances), must be assessed for each plant, family or line in breeding populations in order to accurately identify and select superior plants or families that can be developed into cultivars. Traits of major importance in both tomato and pepper include fruit yield, color, size and shape. Other key traits in both crops include leaf canopy cover (to prevent fruit sunburn), days to maturity, plant size (architecture), and drought and heat tolerance. Depending on the end uses of the fruits, other traits may be of importance. For example, in processing tomatoes, red fruit that is square-round in shape with high soluble solids (Brix) content are desirable, as is determinant plant habit for mechanical harvesting. In fresh market tomato, desired fruit shapes, sizes and color vary by market class (e.g., beefsteak, plum, cherry, heirlooms, etc.) and the plant architecture may range from determinant bush to indeterminate vine-types that are trellised and repeatedly hand-harvested. In pepper, target fruit traits for breeding depend partly on whether the peppers are sweet (non-pungent) or hot (pungent), and if they are used fresh, dried, ground or in processed products. Sweet pepper pod types include bell, pimento, Cuban and squash, while pungent (hot) pepper pod types include cayenne, jalapeno, serrano, ancho, pasilla, New Mexican and others. Fruit colors in pepper can vary from green to red, purple, yellow and orange. Pod types are distinguished by their shape as well as by color, pungency (heat) level, aroma and/or flavor; these categories are used by the pepper industry to identify the correct pepper for the proper (end) product. Climate change is projected to reduce crop yields and increase the frequency of extreme weather events in major food producing regions including the arid western United States. Efficient breeding of crop cultivars that are productive and resilient in a changing climate is needed to address this major threat to food security. The current labor-, cost- and time-intensive manual methods of evaluating (phenotyping) traits in field-grown crop plants are a major bottleneck in plant breeding that slows progress. Efficient use of genomic tools (including high-throughput genotyping) in breeding cannot be implemented at the required scale and speed without a transformational change in approach to in-field trait phenotyping. Development of new technologies and approaches for efficient high-throughput in-field phenotyping methodologies are urgently needed by plant breeders to achieve rapid development of new crop varieties to meet the threats of climate change and provide food security for a growing population. The developmental pace of in-field phenotyping technology lags behind that of genomics technology. As a consequence, trait phenotyping in the field is a significant bottleneck for efficient breeding of crop cultivars. The development and application of technologies that result in accurate, high-throughput plant phenotyping in the field would serve to accelerate breeding crop cultivars. More recent developments in sensors, machine vision and machine learning techniques, higher resolution digital cameras, massively parallel data processing power and other applicable technologies have now paved the way to enable high-throughput in-field plant phenotyping (HTPP) in the field to benefit food crop breeding programs. In this three-year project we will focus on the development and deployment of automated, smart HTPP systems to accelerate breeding of new and novel vegetable cultivars adaptable to climate change. We will focus on tomato and pepper, important members of the Solanaceae family of vegetable crops (i.e., tomato, pepper, eggplant, potato). Tomato and pepper have similarities in breeding objectives for a number of plant and fruit traits of importance to growers, consumers and processors. Longer term (beyond this project timeframe) we plan to adapt the HTPP technologies and methods that we develop for tomato and pepper to other important food crops.
Animal Health Component
33%
Research Effort Categories
Basic
(N/A)
Applied
33%
Developmental
67%
Classification

Knowledge Area (KA)Subject of Investigation (SOI)Field of Science (FOS)Percent
2011469108150%
4027299202050%
Goals / Objectives
Long-term goal:Development and deployment of automated, smart high-throughput in-field plant phenotyping (HTPP) systems to accelerate breeding of new and novel vegetable cultivars adaptable to climate change. In this three-year project we will focus on tomato and pepper, important members of the Solanaceae family of vegetable crops (i.e., tomato, pepper, eggplant, potato). Tomato and pepper have similarities in breeding objectives for a number of plant and fruit traits of importance to growers, consumers and processors. Longer term (beyond this project timeframe) we plan to adapt the HTPP technologies and methods that we develop for tomato and pepper to other important food crops.Supporting objectives of the proposed project: Objective 1. Development of automated, high-throughput plant phenotyping (HTPP) technology for predictive phenotyping of fruit and plant traits relevant to breeding tomato and pepper grown using commercial on-farm cultural practices. Objective 2. Performance assessment and validation of automated HTPP predictive phenotype measures using "ground truth" trait data obtained by plant breeders via manual phenotyping of plant and fruit traits in the field. Objective 3. Initiate deployment of automated HTPP in vegetable breeding by field evaluation of segregating breeding lines genetically unique from those evaluated in years 1 and 2 to assess the ability of HTPP to identify superior breeding lines.
Project Methods
Activities for Objective 1: Automated high-throughput plant phenotyping (HTPP) will be achieved by developing a physical research prototype machine and custom software designed to measure breeding trait phenotypes in tomato and pepper that span a range of plant and fruit characteristics, and include traits of importance to vegetable growers, processors and consumers. To address current phenotyping bottlenecks in breeding progress, our approach is to develop a mobile smart HTTP machine (ground-based) that uses non-contact sensing methods to acquire measurement data while the sensor platform is in motion. In this project, trait phenotypes are classified as either exterior or interior, based upon the location relative to the plant canopy envelope where the required information resides. Because some of the important traits require detailed information about interior traits, our proposed approach is to use proximal sensing techniques from a mobile ground vehicle. HTPP measurements will be achieved using a sensor suite consisting of the existing high-resolution digital cameras of the Robotic in-Field Phenotyping System (RFPS) plus new sensor capability for real-time leaf temperature measurement. Raw sensor data will be analyzed using a multifaceted approach. For many exterior plant phenotypes, algorithms developed in the Slaughter lab and elsewhere will be adapted for application to HTPP data for plants in plot-sized groups once canopy closure within the row occurs. Further adaptations of the algorithms will be implemented to run at a higher-throughput on the high performance computing cluster at UC Davis. For some trait phenotypes where no prior research has been conducted, new algorithms will be developed in this project. For interior phenotypes and some whole plant phenotypes a multi-step, machine learning approach based upon stereology techniques will be used. In this case, phenotype estimates will be generated using an expert system multiclassifier approach. Separate predictive models for each interior phenotype and morphological category will be created to predict the trait phenotype across the whole plot. A stereological plant model for the most probable morphological category will then be used to estimate the phenotype based upon visible information accessible to the RFPS.Activities for Objective 2: Breeding materials for field experiments: Tomato: For our experiments in this project, a set of processing tomato breeding lines segregating for tolerance to abiotic stresses (drought & high temperatures), plant traits (e.g., plant size, plant shape, etc.) and fruit traits (color, size, shape, etc.) will be evaluated in replicated field experiments at UC Davis. Genotyped, advanced tomato breeding lines derived from crossing cultivated tomato with wild tomato species with agriculturally desirable traits, including tolerance to abiotic stresses (e.g., drought) and fruit traits (yield, maturity, etc.) will be studied. Pepper: Advanced pepper breeding lines and cultivars, including New Mexican red chile peppers, yellow wax peppers, and bell peppers will be studied. These lines and cultivars represent a range of plant traits, fruit traits and pod types that are important in pepper. Materials selected for tolerance to reduced irrigation water will be included in our field experiments.Field experiment design and methods: For in-field breeding experiments, a split plot experimental design will be used for each crop each year of this project: main plots will be the drip irrigation treatment with three blocks per main plot, and subplots will be the breeding lines, with border rows between irrigation treatments. The two split plot experiments (one for tomato, one for pepper) will be arranged next to each other in the field, such that all the pepper plots are located in one split plot experiment, and all the tomato plots are in the other split plot experiment. This experimental layout will allow each crop to be managed separately for fertilizer, weeds, pests/diseases, etc. using standard commercial practices for each crop. Seedlings will be hand- transplanted into 10-plant plots on 1.52 m center beds, with 0.3 m between plants within a row at UC Davis. In year 3, we will use the same experimental design but with segregating breeding lines that are genetically unique from those evaluated in years 1 and 2 to assess the ability of HTPP to identify superior breeding lines.Trait phenotyping performed by HTPP and manually by breeders: HTPP measurements will be made in the 10-plant plots at regular time intervals to characterize plant and fruit traits across a range of developmental stages and to allow a time-series analysis of the HTPP measurements. A logging weather station will be used in the experimental field to record weather data. Weather data will be combined with leaf temperature distributions and plant size estimates to estimate abiotic stresses levels. In parallel with HTPP, plant and fruit traits will be evaluated manually in the field by plant breeders and graduate students. Manual phenotyping will be done subjectively for some traits with scoring scales that have been used previously in our breeding programs, and for other traits we will develop appropraite objective measures. A sample of 20 fruit from each subplot will be harvested and weighed, and analyzed for size and shape. Fruit yield will be obtained at season end from two plants per subplot: all fruit will be removed, sorted by color/maturity, counted and weighed, and then two plant shoots without fruit will be weighed to estimate fresh weight plant biomass.Trait data analyses and comparisons to HTPP data: Trait data obtained manually by the breeders will be analyzed on a per-crop and per-trait basis with ANOVA for a split plot design. Effects of the two irrigation treatments in the split plot experiment for each crop will be tested per trait. We will compare trait data obtained manually with the HTPP measurements. For directly comparable measurements, an analysis of covariance will be conducted, with morphological category and irrigation treatment as the covariates, to determine the predictive performance of the HTPP measurements. For categorical phenotypes, a discriminant analysis will be conducted to compare the manual phenotype classifications to the HTPP classifications. For some interior traits, a time series analysis of covariance approach will be taken to determine the performance by morphological category. We will also conduct a comparison of the degree of genotype separation by calculating the mean separations and precision using a variance normalized multidimensional distance measure between the genotype means and the within genotype variance for both the manual method and the HTPP method.Activities for Objective 3: Breeding materials, field experiments, trait phenotyping: Tomato and pepper breeding lines that are genetically unique from those used in the first two years will be employed in year 3 of the project. The field experimental design, field methods, traits and phenotyping methods, and data analyses will be the same as those described under Objective 2. A direct comparison of ranking of breeding lines' trait performance as measured manually and with HTPP, will also be conducted. Spearman's rank correlations will be calculated by crop and by trait for the two phenotyping methods to determine the degree of concordance or agreement. For each crop, the trait data will also be evaluated using principle component analysis (PCA) and cluster analysis to evaluate if a similar set of best performing lines for multiple traits are identified by both HTPP and manual phenotyping.

Progress 02/01/17 to 01/31/21

Outputs
Target Audience:The primary target audience are the stakeholders in the vegetable crops industry who breed new cultivars of vegetable crops, the technology manufacturers and service providers who develop commercial products for and provide services to plant breeders, and the consumers who purchase and consume their products. This group includes vegetable crop farmers, as well as on-farm workers, and individuals and industries who supply goods and services to farmers, all of whom may be affected by the new technologies developed. It also includes the new engineers, plant scientists, and students who will be trained in the development of and use of the new high-throughput plant phenotyping technologies being developed for the vegetable crop industry. The audience also includes engineers and plant scientists external to the project who through the exposure to the results and findings of this project will be enabled to develop companion technologies and synergistic practices and systems that can further advance the long-term sustainability and profitability of specialty crops grown in the USA. Policy makers are also part of the target audience because the technology being developed has the potential to create a transformational change in approach to in-field trait phenotyping and the rate of development of new crop varieties to meet the threats of climate change and provide food security for a growing population. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We were able to train two engineering graduate students in the creation of novel, state-of-the-art software to automate the analysis of automatically collected in-field high-throughput phenotyping data. The engineering graduate students were also trained to use machine learning and image processing techniques to analyze 2D and 3D image data. These students were also trained in advanced methods of analysis for very large databases and the use of modern tools for the use of artifiical intelligence techniques in the anaysis of this data. A plant sciences graduate student was trained in implementing new technology into fields as well as designing and managing the plots while meeting the specific needs of high-throughput plant phenotyping system. Several engineering undergraduate students were exposed to large scale data analysis and implementing sophisticated machine in an agricultural field. The engineering undergraduate students developed automated methods to organize image data into a meaningful format to create a database needed for high-throughput phenotyping. Many plant sciences undergraduate students were exposed to collecting ground-truth data for analysis and verification against the HTPP data. How have the results been disseminated to communities of interest?Multiple presentations were made at the Annual International Meeting of the American Society of Agricultural and Biological Engineers and at the Internaltional Meeting of the Society of Photo-Optical Instrumentation Engineers' technical conference on Autonomous Air and Ground Sensing Systems for Agricultural Optimization and Phenotyping. Both meetings were held virtually due to the pandemic. What do you plan to do during the next reporting period to accomplish the goals? Nothing Reported

Impacts
What was accomplished under these goals? Impact Statement This project addressed the critical need for improved and more efficient methods for collecting in-field phenotyping data in support of breeding of crop cultivars to address critical farming needs such as climate change and drought. In this project, novel technologies and a novel agricultural vehicle for conducting high-throughput in-field phenotyping of important members of the Solanaceae family of vegetable crops (i.e., tomato, pepper) were created and evaluated. Vegetable crop breeders currently make manual and subjective measurements of key traits (called phenotypes). These methods are labor-, cost- and time-intensive, inherently imprecise and provide sparse amounts of information. They are a major bottleneck that slows progress for new cultivar development. The automated technologies and advanced methods developed in this project demonstrate the opportunity for a fundamental change in how vegetable crops breeders assess key traits. Novel automated phenotyping approaches not previously considered by breeders were created that provide new opportunities to assess breeding lines. The project also accomplished a change in knowledge for both engineers and plant breeders in the approach to how breeding trials should be designed to ensure greater efficiency. The efficiency gains by adopting and continuing to develop and understand the new high-throughput phenotyping technologies should lead to further improvements in the efficient breeding of crop cultivars. Objective 1 Accomplishments A novel, automated agricultural machine was created for real-time collection of high-throughput in-field phenotyping data in Solanaceae vegetable crops (i.e., tomato, pepper). This new machine was adapted, improved and customized over the course of the project to specifically serve the needs of vegetable crops breeders. The design used a novel on-board sensor-fusion management system for collection of many types of non-destructive measurements, including 2D color and 2D normalized difference vegetation index images, infrared leaf temperature measurements, 3D time-of-flight measurements of plant architecture, and high-accuracy geospatial location measurements. The design also provided novel techniques for the control of environmental factors that allowed it to collect high-throughput data from a continuously moving vehicle traveling outdoors. A novel, multi-pose sensor design allowed it to collect images of different parts of the vegetable plants. The multi-view data allowed it to phenotype both immature green vegetables nestled among the green canopy and ripening vegetables. Noncontact leaf temperature measurements allowed it to characterize plant water stress. The design automatically created metadata information, identifying the breeding plot where automated phenotyping data was being collected at any moment. Over 20 input data streams were collected when operating in a field. New methods of handling, transfer, organization and storage of data, a critical issue in automated phenotyping, were designed to be geospatially aware and efficient, greatly improving throughput and reducing data storage requirements. The large size of the breeding trial fields used in the study resulted in 1.4 Terabytes of image data collected each day, in addition to the time-of-flight 3D plant architecture and leaf temperature data, for example. A system to combine and catalog data so it could be automatically referenced by genotype, location in the field, and date, allowed breeders to observe how the phenotype changed over the season. The unique nature of the phenotypes of interest to the breeders of the Solanaceae vegetable crops (such as day of appearance of the first immature green fruit) required new algorithms and methods of analysis, which were successfully created. Many machine learning techniques involving neural networks for phenotype identification in Solanaceae vegetable crops were successfully created. Objective 2 Accomplishments The performance of the novel phenotyping system was assessed through a series of in-field breeding trials (tomato and pepper) conducted on a farm. Automated phenotyping data was collected semiweekly. The traditional, manual breeder's phenotypes were collected at their normal single timepoint in the season. Advanced methods were used to process, model and analyze the automated phenotyping data and to characterize its performance in comparison to the manual breeder's phenotypes (traits related to flowers, green fruit, red fruit, and sunburnt fruit, plant size, plant shape). An automated computer method for distinguishing between green fruit and green foliage using pseudo-NDVI images was successfully developed. A very challenging task for humans, who often rely on physically touching and disturbing the plants for manual phenotyping. The automated algorithm utilized machine learning techniques to successfully predict the location of green fruit at very high-resolution (pixel level). An automated computer algorithm was successfully created to identify individual fruits at stages of development using an AI- based neural network. The algorithm was trained on the phenotype data collected by humans. Results showed that the neural network could make similar predictions to humans when viewing the same scene. The algorithm was able to detect a tomato flower while also tracing each individual 1-cm long petal. The algorithm was 96% accurate at automatically detecting objects (flowers, red fruit, sunburnt fruit, etc.). A 2nd network was created for green fruit detection using the pseudo-NDVI images. The accuracy for the NDVI algorithm was 95% at automatically detecting green vegetable fruits. Multi-year, time-of-flight, 3D plant architecture data was collected and analyzed and compared to manual phenotypes (both hand measurements and subjective indices) across 3 years of study. The system was able to successfully create a 3D reconstruction of each plot using automated high-throughput methods. From the 3D reconstruction, architectural phenotypes like height, width, volume and shape index were successfully created for each plot. The automated 3D plant architecture data was compared to the manual phenotypes and characterized. Results showed, for example, that the human assessment of plant height was consistently at the 90th percentile of heights shown in the 3D reconstruction for each genotype of plant across the dates data was collected. Objective 3 Accomplishments An automated high-throughput in-field phenotyping system was successfully deployed in outdoor breeding trials. Similar breeding trials utilizing both tomato and pepper breeding lines selected based upon the previous year's results were used in the final year's trial. Automated high-throughput in-field plant phenotyping sensor data was collected semiweekly. Manual breeder's phenotypes were collected at their normal single timepoint in the season. Advanced methods were successfully used to process the automated high-throughput phenotyping data and to characterize breeding line segregation performance in comparison to the traditional manual method. Comparable levels of segregation of the breeding lines were obtained by the automated system as compared to manual methods. Additional future study is recommended with a more diverse set of genotypes in order to fully explore the opportunities provided by many of the novel phenotypes (such as plant volume, or time series analysis) created by the automated high-throughput phenotyping system.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2020 Citation: Roo, C., D.C, Slaughter, V. Vuong. 2020. Volume Estimation in High Throughput Phenotyping using 3D xBox Kinect Data. ASABE Paper No. 2001502. 2020 Annual International Meeting. Held Virtually. July 13-15, 2020.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2020 Citation: Vuong, V., D.C. Slaughter, C. Roo, D. St. Clair, B. Kubond, P. Bosland. 2020. In-field high-throughput phenotyping approach using a multi-view and multi-sensor ground-based vehicle. Proc. Society of Photo-Optical Instrumentation Engineers (SPIE) 11414, Autonomous Air and Ground Sensing Systems for Agricultural Optimization and Phenotyping V, 114140H (27 April 2020); https://doi.org/10.1117/12.2560531


Progress 02/01/19 to 01/31/20

Outputs
Target Audience:The primary target audience are the stakeholders in the vegetable crops industry who breed new cultivars of vegetable crops, the technology manufacturers and service providers who develop commercial products for and provide services to plant breeders, and the consumers who purchase and consume their products. This group includes vegetable crop farmers, as well as on-farm workers, and individuals and industries who supply goods and services to farmers, all of whom may be affected by the new technologies developed. It also includes the new engineers, plant scientists, and students who will be trained in the development of and use of the new high-throughput plant phenotyping technologies being developed for the vegetable crop industry. The audience also includes engineers and plant scientists external to the project who through the exposure to the results and findings of this project will be enabled to develop companion technologies and synergistic practices and systems that can further advance the long-term sustainability and profitability of specialty crops grown in the USA. Policy makers are also part of the target audience because the technology being developed has the potential to create a transformational change in approach to in-field trait phenotyping and the rate of development of new crop varieties to meet the threats of climate change and provide food security for a growing population. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We were able to train two engineering graduate students in the design and implementation of state-of-the-art software and hardware used to collect phenotypic data as well as the implementation of software to automate the analysis of data. The engineering graduate students were also trained to use machine learning and image processing techniques to analyze image data. A plant sciences graduate student was trained in implementing new technology into fields as well as designing and managing the plots while meeting the specific needs of high-throughput plant phenotyping system. Several engineering undergraduate students were exposed to large scale data analysis and implementing sophisticated machine in an agricultural field.The engineering undergraduate students developed automated methodsto organize image data into a meaningful format to create a database needed for high-throughput phenotyping.Many plant sciences undergraduate students were exposed to collecting ground-truth data for analysis and verification against the HTPP data. How have the results been disseminated to communities of interest?Presentations were made at the Annual Meeting of the National Association of Plant Breeders, the American Society of Plant Biologists Annual Meeting, and teh Plant and Animal Genome Conference. We disseminated our methods and results to plant breeders, engineers, scientists, and people interested in the growth of agriculture in the form of presentations, in-field demonstrations and in-person meetings. What do you plan to do during the next reporting period to accomplish the goals?Currently there are various data analysis methods being developed. Using the information collected from the Time-of-Flight cameras, a canopy volume estimation method is in development. This system takes the 3D pointcloud created by the Time-Of-Flight cameras to estimate canopy volume, height, and width. Software is also being developed for automated fruit and flower counting. This software utilizes machine learning and neural networks to detect flowers and fruit in a given image. This can be used to estimate phenotypes used by plant breeders, such as day of first flower, green fruit, ripe fruit, and sunburnt fruit and total ripe fruit yield. Non-contact temperature data will be analyzed as well. This temperature data will quantify if a non-contact temperature sensor can estimate canopy volume and how many sensors are necessary to get accurate measurements. In 2020 a research field will be planted and used for high-throughput phenotyping. This field will be used to collect data that was not able to be collected in previous years due to inclement weather.

Impacts
What was accomplished under these goals? Under Objective 1 we were able to continue utilizing the automated high-throughput phenotyping system in a plant breeding trial.The control of the driving speed of the high-throughput phenotyping system affected the quality of data collection. In order to ensure high quality and reliable data collection throughout the season a new speed control method was implemented. The system's within breeding plot travel was timed when driving down a row and when the system was at the end of the row the data quality was checked. A desired speed was determined and the plot traveersal time was used to ensure consistent speed and data quality. Automated mapping of the data collection process in real time was important because of the amount of data collected on each day. The automated mapping in real time made it possible to only collect necessary data (no images of empty dirt). The field for the breeding trials used a new irrigation system, which mitigated issues from previous years where the drought treatments were not implemented as homogenously as desired for ground-based high-throughput phenotyping.Under Objective 2 we were able to achieve our high-throughput objective to collect data at the throughput level proposed, and were able to completely measure all plants in three separate breeding trial fields (two of tomatoes and a field of peppers) in under 3 hours in 2019.From the 2018 data we were able to create a database using the images collected. The database allows researchers to access the images of the tomato plant throughout the growing season at all different camera angles. Given the throughput, vehicle travel speed and noninvasive and noncontact nature of our Field Phenotyping System, these results are considered very good for HTPP in a horticultural crop like tomato. In comparison, the existing manual method is invasive and extremely time consuming. A replicated field trial was conducted in 2019 using two Solanaceae crops, tomatoes and peppers. The field consisted of 24 tomato cultivars and 24 pepper cultivars. The field was phenotyped using the high throughput system and manual phenotyping.The field was asssessedthe ability of the high-throughput phenotyping system to identify superior breeding lines, using the manual phenotyping as the ground truth.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2019 Citation: Slaughter, D., St. Clair, D., Bosland, P., Vuong, V., Kubond, B., Roo, C. 2019. High-Throughput In-Field Phenotyping Systems to Accelerate Breeding of Climate-Resilient Vegetable Crops. Presentation at the 2019 Annual Meeting of the National Association of Plant Breeders. Pine Mountain, GA.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2019 Citation: Slaughter, D.C., 2019. Farm Forward  the future of Smart Farming Technologies in America. Plenary symposium called Future Food and Agriculture at the American Society of Plant Biologists (ASPB) annual meeting. San Jose, CA.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2020 Citation: Kubond, B., Vuong, V., Kaur, A., Bosland, P., Slaughter, D.C., St. Clair, D.A. 2020. PE0068 Evaluation of Tractor-Based High Throughput Phenotyping Methods for Use in Tomato and Pepper Breeding. Plant and Animal Genome XXVIII Conference January 11-15, 2020


Progress 02/01/18 to 01/31/19

Outputs
Target Audience:The primary target audience are the stakeholders in the vegetable crops industry who breed new cultivars of vegetable crops, the technology manufacturers and service providers who develop commercial products for and provide services to plant breeders, and the consumers who purchase and consume their products. This group includes vegetable crop farmers, as well as on-farm workers, and individuals and industries who supply goods and services to farmers, all of whom may be affected by the new technologies developed. It also includes the new engineers, plant scientists, and students who will be trained in the development of and use of the new high-throughput plant phenotyping technologies being developed for the vegetable crop industry. The audience also includes engineers and plant scientists external to the project who through the exposure to the results and findings of this project will be enabled to develop companion technologies and synergistic practices and systems that can further advance the long-term sustainability and profitability of specialty crops grown in the USA. Policy makers are also part of the target audience because the technology being developed has the potential to create a transformational change in approach to in-field trait phenotyping and the rate of development of new crop varieties to meet the threats of climate change and provide food security for a growing population. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We were able to train an engineering graduate student in the design of state of the art technology used to collect phenotypic data as well as the implementation of software to automate the analysis of the data. A plant sciences graduate student was trained in implementing new technology into fields as well as designing and managing the plots while meeting the specific needs of high-throughput plant phenotyping system. Several engineering undergraduate students were exposed to large scale data analysis. Many plant sciences undergraduate students were exposed to collecting ground-truth data for analysis and verification against the HTPP data. How have the results been disseminated to communities of interest?The results have been disseminated to communities such as engineers, plant breeders, and people interested in the growth of agriculture in the form of presentations, in-field demonstrations and in-person meetings. Presentations were given at the ASABE Annual International Meeting, Detroit, MI, USA and to various national and regional industry groups holding annual meetings in California. What do you plan to do during the next reporting period to accomplish the goals?For the next reporting period we plan to add thermal sensors to collect leaf temperature data for drought related breeding target phenotypes and deploy a weather station to monitor the weather. We also plan on optimizing software and data turnover time to increase the high-throughput aspect. We will create custom software to produce HTPP prections for the remaining breeding target phenotypes. Finally, another breeding trial will be conducted to address work proposed under Objective 3 and to prove the efficiency of the system.

Impacts
What was accomplished under these goals? Under Objective 1 we successfully modified the automated Field Phenotyping Vehicle to improve the quality of the image data collected for late-season breeding target traits for high-throughput plant phenotyping (HTPP) in tomato and pepper. The design changes accomplished in this period were significant and required a complete change in our multi-view approach to better accommodate the large diversity in plant architectures in tomato and pepper breeding trials of this type. The camera array was re-oriented and the central arch moved to be further away from the plants and equidistant from the center of the row. A nominal distance of approximately 90 inches from plant bed to camera sensor was identified as the optimum to achieve the needed depth of focus and illumination uniformity, while still allowing the Field Phenotyping Vehicle to enter the traditional row spacing of commercial tomato production. White corrugated twin-wall plastic covered the entire sensor module to create a rigid structure resistant to wind as well as to provide even lighting at all times of the day. Supplemental white and infrared LED lighting modules were embedded in the camera array to increase light levels and allow real-time data capture at higher travel speeds. An additional time-of-flight camera module was added to the top of the sensor module, looking down at the plants, to get a third view of the plants and improve the quality of the systems ability to predict breeders' plant architecture phenotype targets. Additional sensors and control systems were incorporated to further improve the automation and high-throughput of the entire process. The improvements added to the Field Phenotyping Vehicle in this project year were focused at optimizing data collection and were successful at reducing post-processing times. Under Objective 2 we were able to achieve our high-throughput objective to collect data at the throughput level proposed, and were able to completely measure all plants in three separate breeding trial fields (two of tomatoes and a field of peppers) in under 3 hours in 2018. From the 2017 in-field HTTP data we were able to create a model that could predict the genotypes that produce the most fruit yield at the end of the season. Compared to the ground-truth data we were able to get 70% accuracy. Given the throughput, vehicle travel speed and noninvasive and noncontact nature of our Field Phenotyping System, these results are considered very good for HTPP in a horticultural crop like tomato. In comparison, the existing manual method is invasive and extremely time consuming. Three-dimensional plant architecture data collected automatically by our Field Phenotyping System was also analyzed to predict the plant height and width of different genotypes.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David C. Slaughter. 2018. COMMUNICATION, LITERACY, EDUCATION FOR AGRICULTURAL RESEARCH (CLEAR), ROBOTS AND DRONES: INNOVATIONS IN AGRICULTURAL RESEARCH. BERKELEY, CA, USA 04/03/2018.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David C. Slaughter, 2018. CALIFORNIA LEAGUE OF FOOD PROCESSORS (CLFP) AG PRODUCTION COMMITTEE, SPRING MEETING. LATEST ADVANCEMENTS IN NEW TECHNOLOGIES FOR AUTOMATION IN ON-FARM TOMATO PRODUCTION. FRESNO, CA, USA 3/13/2018
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David C. Slaughter, 2018. PRECISION AG PANEL FOR THE CAST: COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY BOARD OF DIRECTORS ANNUAL MEETING. SMART FARM: CREATING THE FARM AND FARM WORKERS OF THE FUTURE. SACRAMENTO, CA, USA, 10/24/2018.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David C. Slaughter, 2018. CHERRY DAY ANNUAL MEETING. FARM FORWARD, CREATING THE SMART FARM OF THE FUTURE. STOCKTON, CA, USA. 03/07/2018,
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: David C. Slaughter, 2018. AMERICAN SEED TRADE ASSOCIATION (ASTA). SPECIAL SEED CENTRAL EVENT: "ASTA MEETS UC DAVIS". FARM FORWARD, CREATING THE SMART FARM OF THE FUTURE. DAVIS, CA, USA. 10/04/2018,
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Vivian L Vuong, David C Slaughter, Dina St. Clair, Paul Bosland, Bryce Kubond, Amanjot Kaur. 2018. HIGH-THROUGHPUT PHENOTYPING METHODS FOR GREEN FRUIT. ASABE Annual International Meeting, Detroit, MI, USA  07/30/2018
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Dina St. Clair. 2018. Processing Tomato Conference, BREEDING CLIMATE-RESILIENT TOMATO FOR CALIFORNIA USING WATER STRESS-TOLERANT WILD TOMATO, Davis, CA, USA, 12/13/18
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Dina St. Clair. 2018. CLFP Agricultural Production Committee Meeting, EXPLORING THE WILD TOMATO GENOME FOR BREEDING WATER STRESS TOLERANCE, Merced, CA, USA, 11/2/18
  • Type: Conference Papers and Presentations Status: Published Year Published: 2018 Citation: Dina St. Clair. 2018. New Mexico Chile Pepper Conference, HIGH THROUGHPUT PHENOTYPING PROJECT WITH CHILE PEPPERS, Las Cruces, NM, USA, 2/6/18


Progress 02/01/17 to 01/31/18

Outputs
Target Audience:The primary target audience are the stakeholders in the vegetable crops industry who breed new cultivars of vegetable crops, the technology manufacturers and service providers who develop commercial products for and provide services to plant breeders, and the consumers who purchase and consume their products. This group includes vegetable crop farmers, as well as on-farm workers, and individuals and industries who supply goods and services to farmers, all of whom may be affected by the new technologies developed. It also includes the new engineers, plant scientists, and students who will be trained in the development of and use of the new high-throughput plant phenotyping technologies being developed for the vegetable crop industry. The audience also includes engineers and plant scientists external to the project who through the exposure to the results and findings of this project will be enabled to develop companion technologies and synergistic practices and systems that can further advance the long-term sustainability and profitability of specialty crops grown in the USA. Policy makers are also part of the target audience because the technology being developed has the potential to create a transformational change in approach to in-field trait phenotyping and the rate of development of new crop varieties to meet the threats of climate change and provide food security for a growing population. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We were able to train an engineering graduate student in the design of state of the art technology used to collect phenotypic data as well as the implementation of software to automate the analysis of the data. A plant sciences graduate student was trained in implementing new technology into fields as well as designing and managing the plots while meeting the specific needs of high-throughput plant phenotyping system. Several engineering undergraduate students were exposed to large scale data analysis. Many plant sciences undergraduate students were exposed to collecting ground-truth data for analysis and verification against the HTPP data. How have the results been disseminated to communities of interest?The results have been disseminated to communities such as engineers, plant breeders, and people interested in the growth of agriculture in the form of poster presentations, oral presentations, and web articles. Poster presentations include the National Association of Plant Breeders. Oral presentations and demonstrations include the California Seed Association and the World Food Center. What do you plan to do during the next reporting period to accomplish the goals?The plan for the next research period is to update the design of the HTPP to improve the quality of data as well as the high-throughput aspect. This will be accomplished by adding geospatial control to reduce redundancy and adding sensor technology to improve the ability to find unripe fruit as well as collect temperature data. Finally, a breeding trial will be conducted to prove the efficiency of the system.

Impacts
What was accomplished under these goals? Under Objective 1 we accomplished designing, building, and testing an automated high-throughput plant phenotyping system. Under Objective 2 were able to collect data at the throughput level desired, able to do an entire field of tomato and pepper crops in 4 hours. However, the implementation of real-time geospatial control is a critical aspect of reducing unnecessary data in the validation. Improvements in the optical geometry of the phenotyping technology is an important aspect in facilitating the automation of analysis, the level of redundancy in the 2017 design may not be necessary to further achieve the project objectives.

Publications

  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: David C. Slaughter, Dina A. St.Clair, Paul W. Bosland, Thuy T. Nguyen, Vivian L. Vuong, Bryce A. Kubond, and Amanjot Kaur. 2018. HIGH-THROUGHPUT IN-FIELD PHENOTYPING SYSTEM TO ACCELERATE BREEDING OF VEGETABLE CROPS. National Association of Plant Breeders (NAPB) Annual Meeting, Davis, CA, USA, August 7-10, 2017. https://napb2017.ucdavis.edu/
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: David C, Slaughter. HIGH-THROUGHPUT IN-FIELD PHENOTYPING SYSTEM TO ACCELERATE BREEDING OF VEGETABLE CROPS. West Coast Food Summit. May 11, 2017. Davis, CA, USA.
  • Type: Websites Status: Published Year Published: 2017 Citation: Andy Fell. Creating the Farm and Farmworkers of the Future. January 24, 2018 in Food & Agriculture. https://www.ucdavis.edu/news/smart-farm
  • Type: Other Status: Other Year Published: 2017 Citation: David C. Slaughter, Thuy T. Nguyen, Vivian L. Vuong. HIGH-THROUGHPUT IN-FIELD PHENOTYPING SYSTEM TO ACCELERATE BREEDING OF VEGETABLE CROPS. Davis Chancellor's Club Insider's Tour, Precision Farm Initiative, Sustainable Farming for the Future. October 7, 2017. https://giving.ucdavis.edu/recognition-resources/events
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: David C, Slaughter. Innovations and new technologies in California Agriculture. California Seed Association 77th Annual Convention. San Diego, March 5-8, 2017.
  • Type: Conference Papers and Presentations Status: Published Year Published: 2017 Citation: David C, Slaughter. Innovations in agricultural technology. The Seeds of Our Future: AgTech and the Connected World. Silicon Valley Forum April 3 - 6, 2017, Davis, CA, USA.