XooNet: A High-Throughput UAV-Based Approach for Field Screening of Bacterial Blight (BB)-Resistant Germplasm in Wild Rice

Pan Pan¹WenLong Guo²MingXia Li¹Haochun Li¹Jingxi Yang¹Zhihao Guo¹Huibo Zhao³Guoping Yu³Maomao Li⁴Long Yi⁴Xiaoming Zheng⁵Guomin Zhou⁶Jianhua Zhang^1*

• ¹Agricultural Information Institute of CAAS, Beijing, China
• ²China Agricultural University College of Agronomy and Biotechnology, Beijing, China
• ³China National Rice Research Institute, Hangzhou, China
• ⁴Jiangxi Academy of Agricultural Sciences, Nanchang, China
• ⁵Chinese Academy of Agricultural Sciences, Beijing, China
• ⁶Nanjing Institute of Agricultural Mechanization Ministry of Agriculture and Rural Affairs, Nanjing, China

Bacterial blight (BB) poses a significant threat to rice production, necessitating efficient screening of resistant wild rice germplasm to facilitate breeding. Traditional methods are labor-intensive and subjective, while existing UAV-based approaches suffer from high costs or incomplete solutions. This study introduces XooNet, a novel UAV-based method for automated BB resistance screening in wild rice, which classifies wild rice into several levels based on BB resistance. To facilitate this method, a high-precision and lightweight oriented bounding box (OBB) detection algorithm for BB in wild rice has been developed. Experimental results show that the screening method achieved an accuracy of 97.5%. After applying the LAMP pruning strategy to balance performance and efficiency, the detection model achieved an accuracy of 93.1% with a significantly reduced parameter size of 1.4M and a computational complexity of 3.5 GFLOPs. This approach will facilitate the high-throughput screening of extensive wild rice germplasm for BB resistance, thereby expediting the discovery of valuable wild rice genetic resources.