Explainable Deep Learning-Based Image Analysis with Blackbird RGB Imaging Robot for Laboratory High Throughput Phenotyping

Goals / Objectives
Laboratory computer vision is being used to quantify visible traits from images. ARS and Cornell have amassed expertise in high throughput phenotyping for grapevine powdery mildew severity via automation and machine learning data analysis. The goal of this project is to adapt current machine learning data analysis methods to other traits and containerize these for distribution via ARS high performance computers. This will empower additional ARS scientists for user-friendly and effective high-throughput lab phenotyping.

Project Methods
1) Provide two imaging robots to collaborating ARS laboratories for testing new crops and/or traits. 2) Guide those labs through imaging and data analysis. 3) With the SCINet team, design an optimal solution for data uploading, curation, and management. 4) Develop training materials to facilitate widespread deployment of Blackbird in ARS and university labs as a uniform phenotyping platform. 5) Refine the developed cloud-based image analysis system for Blackbird imaging robots 6) Establish a multimodal dataset with human annotation for deep learning model training 7) Explore the use of cutting-edge deep neural networks with the multimodal training dataset for improved phenotypic trait extraction and explanation, especially time-series image datasets for dynamic phenotypes.