An algorithm for target detection in the Bai-yang lake based on DMW-YOLOv8

PENG Kaiyuan; TIAN Liqin

doi:10.19956/j.cnki.ncist.2025.03.007

Abstract：The Bai-yang Lake,known as the "Pearl of North China," is the largest freshwater wetland system in the North China Plain.The unmanned aerial vehicle(UAV)technology coupled with object detection algo-rithms were used to enhance biological monitoring and conservation efforts in Baiyangdian Wetland,and to scientifically evaluate species'habitat quality,determine protection levels,and formulate corresponding conser-vation strategies,.ased on our comprehensive investigation,we have developed DMW-YOLOv8 as an advanced evolution of the YOLOv8n algorithm.By using the WIoU loss function,the similarity between predic-ted and actual targets is measured more accurately,improving detection accuracy,especially in multi-object scenarios.The MLCA attention mechanism is introduced,which integrates channel and spatial information,as well as local and global information,to enhance the network's representation ability.Additionally,the C2fDCNv4 operator is embedded to optimize the sparse DCN operator,improving practical efficiency.The pro-posed algorithm achieves superior performance metrics,with mAP50 and mAP50-95 reaching 89.4％and 52.0％respectively-representing improvements of 2.7％and 2.2％over the baseline YOLOv8n model.The system demonstrates remarkable computational efficiency,processing at 177.8 FPS(71.2 FPS faster than the original implementation).These advancements make our solution particularly valuable for ecological monitoring applications in the Baiyangdian wetland environment.