Location: Genetic Improvement for Fruits & Vegetables Laboratory
Project Number: 8042-21220-260-001-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 15, 2023
End Date: Sep 14, 2025
Objective:
High throughput technologies are needed to facilitate disease screening and development of disease resistant strawberry cultivars. Foundational components will be built to develop an image-analysis and machine learning system to determine anthracnose crown rot resistance/susceptibility of individual GPS-positioned breeding plants: 1) create a digital map to locate and assign an identity to each strawberry plant in the seedling field; 2) determine the boundaries of each live plant using image analysis and machine learning; 3) determine through both imaging and in-person visual analyses the live/dead = resistant/susceptible status of each plant; 4) use in-person visual analyses to inform image analyses through machine learning; 5) create an analyses output file in an Excel format usable for an existing ARS genetic analyses data pipeline.
Approach:
ARS will establish strawberry breeding seedling fields of approximately 6,000 breeding plants to be analyzed each fall after planting in late July. The cooperator will use robotic mapping and navigation algorithms to create a digital map of the strawberry fields and relate them to existing shape files of the farm boundaries, using real-time kinematic (RTK) positioning available on-farm. The digital map will locate and assign an identity to each of the strawberry plants in the field. ARS will confirm the accuracy of this step and provide corrective information. The cooperator will use image analyses and machine learning to determine the boundaries and the resistance/susceptibility status of each plant. ARS will determine visually the resistance/susceptibility status of each plant to inform the machine learning process. ARS and the cooperatorwill jointly construct the appropriate input parameters and data output format for use in existing ARS genetic analysis methods to determine the number of genes segregating in each breeding family and their mode of action, and to identify breeding parents transmitting resistance genes even if the parents themselves are susceptible. Those breeding parents will be used by ARS more heavily in next year’s crossing plan. The disease of interest for this project is anthracnose crown rot which primarily results in complete plant death. The foundational components developed to create the image-analysis and machine learning system needed for anthracnose crown rot, using simple RGB images, can then be applied to diseases that are more difficult to diagnose using simple visual analyses. For these more challenging diseases, multi-spectral image analysis can be used to generate vegetation index output to the same genetic analysis pipeline used for anthracnose crown rot.
