基于YOLOv11-ACW的野外中草药目标检测研究OA

Objective To systematically address the challenges in wild Chinese herbal medicine detection,including cluttered backgrounds,weak features of small targets,and missed detection due to plant occlusion,this work integrates lightweight feature enhancement techniques with optimization concepts for target detection algorithms.Methods Using a self-built dataset,the YOLOv11-ACW model is constructed based on the YOLOv11n baseline.Specifically,the lightweight multi-scale attention module L-AIFI replaces the SPPF structure to enhance multi-scale feature modeling and target extraction capability in complex backgrounds.A novel CA-Wavelet Pool wavelet pooling unit is designed,which combines coordinate attention with frequency-domain analysis to preserve fine details while suppressing background noise during downsampling.In addition,CARAFE upsampling is introduced to optimize the neck network,dynamically restoring edge textures and alleviating feature distortion caused by conventional upsampling operations.Results Experiments are conducted on the self-built Chinese herbal medicine dataset.The improved YOLOv11-ACW algorithm achieves an mAP@0.5 of 83.8%,representing a 2.7%improvement over the YOLOv11n baseline.The GPU inference speed reaches 242 FPS with only 2.78 M parameters.After deployment on an embedded platform,the inference speed maintains 46 FPS without significant accuracy loss.Conclusion The improved algorithm based on YOLOv11-ACW achieves an excellent balance among detection accuracy,lightweight design,and real-time performance.It can effectively address the key difficulties in wild Chinese herbal medicine detection and meet the requirements for real-time detection in practical scenarios.