Recipient Organization
NORTH CAROLINA STATE UNIV
(N/A)
RALEIGH,NC 27695
Performing Department
(N/A)
Non Technical Summary
Plant diseases cause 20 to 40% of global crop loss annually, resulting in estimated economic losses of $30 billion to $50 billion annually. Precise plant disease phenotyping is crucial to selecting plants in resistance breeding programs for developing better disease-management strategies. However, traditional selection practices for desirable phenotype related to plant diseases and yield is in-field manual scouting, a labor-intensive, subjective, and inefficient process which considerably limits experiment scale and data acquisition frequency. We propose a dynamic in-field phenotyping system for monitoring disease-related traits across multiple scales and environments to expedite genomic research. In this project, we will use tomato production as our model system for optimization, tomato is a major industry in the southeastern United States.
Animal Health Component
40%
Research Effort Categories
Basic
30%
Applied
40%
Developmental
30%
Goals / Objectives
The overarching goal of this project is to enable future comprehensive field-based genomic research with broad applicability for monitoring disease-related traits in cropping systems. This will be accomplished by developing a high-throughput phenotyping system with unique multi-scale sensing capabilities and robust machine learning tools. In this particular application, the system will be optimized for tomato foliar disease, although it could be readily adapted for other vegetables. We will accomplish our goal by completing four specific aims: (1) Design a novel ground robot for side-view and inside-canopy imaging using reinforcement learning in digital twins to optimize robot control. (2) Develop accurate and efficient tools for disease identification and scoring based on UGV-collected data using a multi-task vision attention-based transformer. (3) Enhance heterogeneous coordination of UAV and UGV systems through multi-sensor co-registration to improve disease development course assessment at larger scales. (4) Conduct field validation and comparative analysis of yield performance and disease scoring for crop improvement.
Project Methods
The project aims to create a robotic phenotyping system, HuskyBot2.0, equipped with advanced imaging capabilities and optimized control for navigating through tomato crop rows. The system integrates a robotic arm with a plant endoscope designed for side-view and inside-canopy imaging. These capabilities enable the robot to stop at each plant, capture detailed 3D images, and move efficiently to the next target. This process is enhanced by a sim-to-sim digital twin environment that employs deep learning to refine the robot's trajectory based on previous imaging data, optimizing the path for efficient data collection and yield estimation.For disease identification and scoring, a Multi-task Vision Attention-based Transformer (MVAT) will be developed using images from a curated tomato disease dataset. This dataset will support the training of deep learning models to recognize and score disease severity, adapting to the challenges of variable image characteristics and backgrounds found in natural settings.In parallel, coordination between Unmanned Aerial Vehicles (UAVs) and HuskyBot2.0 will be enhanced to assess disease development on a larger scale. UAVs will capture multispectral and polarimetric data which, when integrated with ground-truth data from the UGV, will improve disease scoring accuracy. A data management system will be set up to handle, store, and analyze the data from both UAV and UGV, facilitating large-scale disease management.Field validation will be conducted to test the system's effectiveness in real-world conditions. Experiments will involve multiple tomato genotypes and disease types, assessing the robotic system's performance in disease detection and yield estimation against manual observations. The goal is to demonstrate the system's reliability and accuracy in improving crop management and disease treatment strategies.