Research on a Multi-Stage Multi-Attention Fusion Approach for Rice Pest Detection

Rapid and accurate identification of rice pests is crucial for ensuring food security. Addressing the insufficient feature extraction capability of traditional YOLO in detecting small rice pests, this paper proposes a staged triple attention fusion object detection method based on the YOLOv8 framework. A triple attention module is constructed comprising Efficient Channel Attention (ECA), Simultaneous Attention Modulation (SimAM), and Global Context Blocks (GC). Adaptive weight fusion and learnable residual scaling enhance feature expression and training stability. Simultaneously, based on the semantic and spatial distribution characteristics of features at different scales, partial or complete triple attention is embedded in stages at shallow, intermediate, and deep layers. This approach highlights local textures while capturing global context, all while controlling computational overhead. Experimental results on the rice pest dataset demonstrate that the improved model significantly outperforms the original YOLOv8 in metrics such as Precision, Recall, and mAP. It effectively enhances the identification accuracy of rice pests in complex field scenarios, providing an efficient and reliable technical solution for intelligent agricultural pest monitoring systems.