Automatic Scoring of Rhizoctonia Crown and Root Rot Affected Sugar Beet Fields from Orthorectified UAV Images Using...

Area of Science:

• Plant pathology
• Agricultural engineering
• Computer science

Background:

• Rhizoctonia crown and root rot (RCRR) causes significant sugar beet yield and quality losses.
• Traditional disease scoring relies on manual expert evaluation, which is time-consuming, subjective, and prone to bias.
• Developing resistant varieties is a key control strategy, necessitating accurate phenotyping in breeding trials.

Purpose of the Study:

• To develop and validate an automated workflow for detecting and scoring RCRR in sugar beet breeding trials.
• To leverage unmanned aerial vehicle (UAV)-based imagery and machine learning for precise disease assessment.
• To compare the performance of different AI models and image analysis strategies for RCRR scoring.

Main Methods:

• A workflow combining red-green-blue (RGB) and multispectral imagery from UAVs was employed.
• Georeferenced images were annotated, and convolutional neural networks (CNNs) were trained to score individual diseased plants.
• Machine learning models, including random forest and k-nearest neighbors, were used to score entire plots, incorporating individual plant data.

Main Results:

• Custom-trained CNNs and pretrained MobileNet achieved high precision (0.73-0.85) in scoring individual plants for RCRR.
• A combined random forest and k-nearest neighbors pipeline demonstrated the best weighted precision (0.67) for plot-level scoring.
• Integrating individual plant scores significantly improved the accuracy of UAV-based automated monitoring for RCRR.

Conclusions:

• The developed workflow provides a reliable, automated method for detecting and scoring RCRR using aerial imagery.
• AI-powered phenotyping offers a standardized and accurate alternative to manual expert scoring in plant breeding.
• Automated monitoring of RCRR using UAVs and machine learning can enhance disease management and breeding efficiency.