To achieve real-time and accurate measurement of clothing thermal resistance,a YOLOv8-based personnel cloth-ing detection model(PCT_YOLO)is developed.The model addresses the issues of poor accuracy and lack of real-time performance when differences between personnel clothing categories are subtle and targets are occluded.Firstly,the Attentional Fast Spatial Pyramid Pooling(ASPPF)module is used to highlight unoccluded key features,thereby reducing the impact of occlusion.Secondly,a feature enhancement fusion module(GS_C2f)is proposed to obtain richer feature representations,making category boundary in-formation more distinct and effectively distinguishing the target from the background.Finally,a lightweight convolutional network(LCN)is employed to enhance the model's real-time capabilities.Experimental results show that on the self-made dataset,PCT_YOLO's achieved an of 93.1%mAP50 and a Parameter(106)count of 2.505×106.Compared to the original YOLOv8n model,the mAP50 increased by 3.0 percentage point and the number of parameters was reduced by 5.05×105.This method fulfills the re-quirement of obtaining clothing thermal resistance through clothing detection in real-world scenarios.
万方数据