A Study on Machine Vision-Based Detection of Rice Diseases and Pests

This study addresses the persistent challenges of low detection accuracy and slow processing speed in rice leaf disease and pest detection by proposing an enhanced computational model built upon the YOLO framework. The proposed architecture incorporates Adaptive Downsampling (ADown) for effective feature extraction, while integrating a Context Anchor Attention (CAA) mechanism that strengthens central region characteristics through global average pooling combined with one-dimensional bar convolution operations. Additionally, a Task Alignment Dynamic Detection Head (TADDH) is implemented to significantly improve small-object recognition capabilities and classification performance. Empirical evaluation demonstrates substantial improvements across key performance metrics, including precision, recall, and mean average precision (mAP). These findings establish the model's capability to provide robust technical support for early-stage detection and precision management of rice foliar diseases and insect pests.