Screening elite bacterial blight (BB) resistant wild rice germplasm for breeding: An AI-driven approach

Bacterial blight (BB), caused by Xanthomonas oryzae pv. oryzae (Xoo), is a devastating rice disease worldwide. Wild rice serves as an invaluable reservoir of BB resistance genes, and introgressing them into cultivated varieties provides a cost-effective, environmentally sustainable strategy for control. Automated, non-destructive, and low-cost screening of BB-resistant wild rice germplasm is crucial for breeding BB-resistant rice varieties. This study proposes an AI-driven approach for screening elite BB-resistant wild rice germplasm to support breeding efforts. An AI-driven workflow for BB resistance screening in wild rice is developed, enabling in-field, repeated evaluations, and is implemented in a portable backpack apparatus costing approximately 890 USD. A computer vision algorithm is designed to segment lesions, measure lesion lengths, and evaluate BB-resistance levels, thereby screening elite BB-resistant wild rice. The lesion segmentation model, with only 0.85M parameters and 4.33G FLOPS, achieves an IoU of 94.14%. Comprehensive evaluations, including BB resistance level evaluation, field testing, and generalization testing, demonstrated high accuracy (92.0, 94.0, and 95.6% respectively) and efficiency gains. This method will support high-throughput screening of large-scale wild rice germplasm for BB resistance, accelerating the exploration of wild rice germplasm.