基于改进RT-DETR水稻病虫害检测模型的研究OA

To address the dual demands for high accuracy and adaptability to complex scenes in rice pest and disease identification tasks,this paper proposes an object detection method based on an improved RT-DETR.Its performance is validated and comparatively analyzed through systematic experiments.Building upon the original RT-DETR model,we incorporate a P2 small object detection head,OrthoNets orthogonal channel attention mechanism,and Additive Block module to enhance the model's capabilities in small object detection and complex background handling.Experimental results show that the improved model achieves a 1.2％increase in accuracy and a 2.3％improvement in mAP@50 compared to the original RT-DETR,with the parameter count reduced from 20.18M to 19M.Despite a 36.5G increase in FLOPs,the detection performance is significantly enhanced.Additionally,the improved model is compared with mainstream object detection algorithms YOLOv5 and YOLOv8,evaluated across multiple dimensions including precision,recall,parameter count,and computational complexity.Results demonstrate that the improved model outperforms the comparison algorithms in key metrics,exhibiting superior robustness and recognition capabilities particularly in small object detection and complex background adaptation.Visual analysis of representative samples further validates the method's effectiveness in practical application scenarios.This research provides an effective technical pathway for intelligent monitoring and precise prevention of rice pests and diseases,laying a foundation for subsequent model deployment and popularization.