一种烟草病虫害混合专家检测变换模型OA

To solve the challenge of multi-scale object detection in tobacco pest and disease identification under complex field surroundings,a mixture-of-expert detection transformer model is proposed.The model in-tegrates three heterogeneous expert modules according to the different feature processing requirements of to-bacco pest and disease images.At the shallow layer,a cross-stage detail enhancement module is used to strengthen local details and edge features;at the middle layer,an adaptive balancing module is utilized to intel-ligently balances local and contextual information;at the deep layer,a cross-stage efficient state-space mod-ule is designed to effectively model long-range dependencies,thereby achieving hierarchical and dynamic ex-traction of multi-scale pest and disease features.Experimental results show that compared with the baseline model RTDETR-R18,the proposed mixture-of-experts detection transformer model reduces parameters and floating-point operations by 27.4％and 16.3％,respectively,while improving mAP@0.50 from 0.770 to 0.815.