Accelerating Wild Rice Disease-Resistant Germplasm Exploration: Artificial Intelligence (AI)-Powered Wild Rice Blast Disease Level Evaluation and Disease-Resistance Identification

Accurate evaluation of disease levels in wild rice materials and disease-resistance identification are critical for developing rice varieties resistant to blast disease. However, existing evaluation methods face limitations that hinder progress in breeding. To address these challenges, we propose an artificial intelligence (AI)-powered method for evaluating blast disease levels and resistance identification in wild rice. A lightweight segmentation model for diseased leaves and lesions was developed, incorporating an improved federated learning approach to enhance robustness and adaptability. Based on the segmentation results and resistance identification technical specifications, wild rice materials were evaluated into 10 disease levels (L0 to L9), further enabling disease-resistance identification by multiple replicates of the same materials. The method was successfully implemented on augmented reality glasses for real-time, first-person evaluation. Additionally, high-speed scanner and edge computing devices were integrated to enable continuous, precise, and dynamic evaluation. Experimental results demonstrate the outstanding performance of the proposed method, achieving effective segmentation of diseased leaves and lesions with only 0.22 M parameters and 5.3 G floating-point operations per second (FLOPs), with a mean average precision ([email protected]) of 96.3%. The accuracy of disease level evaluation and disease -resistance identification reached 99.7%, with a practical test accuracy of 99.0%, successfully identifying three highly resistant wild rice materials. This method provides strong technical support for efficiently identifying materials resistant to wild rice blast disease and resistance breeding efforts.