基于FDP-YOLO的复杂环境下苹果蛾虫害检测方法OA

There are many difficulties in detecting apple moth pests in complex environments:the small size of the targets leads to a lack of features,severe occlusion of pests causes incomplete targets,and variable lighting affects imaging quality.At the same time,practical deployment has high requirements for model lightweight.To address the above issues,this study proposed a lightweight high-precision object detection model,FDP-YOLO.Firstly,a fine-grained channel atten-tion cross-stage fusion mechanism was designed in the backbone network to enhance the feature expression ability of small targets under complex lighting conditions through the interaction of global and local features.Secondly,a dynamic convolu-tion method based on the idea of hybrid experts was introduced in the neck network to improve the feature fusion effect in complex scenes under the premise of main-taining lightweight.Finally,the PIoU loss function was adopted in the detection head to optimize the regression performance of difficult samples under complex conditions such as severe occlusion.Experimental results show that FDP-YOLO maintained lightweight characteristics while the detec-tion accuracy improved by 2.9 percentage points,the recall improved by 1.5 percentage points,and the mAP0.5(mean aver-age precision at an intersection over union threshold of 0.5)improved by 1.7 percentage points compared to the baseline model YOLOv8n.Its mAP50 was significantly better than models such as YOLOv10n and RT-DETR-R50,effectively valida-ted the technical advantages in detection performance of small targets,occluded targets,and variable lighting scenes in complex environments.It can provide technical support for intelligent monitoring of pests in complex agricultural scenes.