四川大学学报（自然科学版）2025，Vol.62Issue(2)：369-376,8.DOI:10.19907/j.0490-6756.240297
基于深度学习的门机运行工况人车检测算法研究
Research on personnel-vehicle detection algorithm for gantry crane working conditions based on deep learning
摘要
Abstract
Hydropower stations generally ensure the safety of the operation of the gantry crane through visual inspection.In the working area of the gantry crane,personnel and vehicle are the main sources of danger.In order to solve the problem of different scales and occlusion of personnel and vehicle targets in the detection im-age,the YOLOv8s model was improved based on deep learning.The backbone was enhanced by substituting C2f with DyEMA_C2f,which integrates EMA and dynamic convolution.This modification improves the model's capacity to learn spatial and channel relationships,enhances feature extraction capabilities for targets of varying scales,and reduces computational overhead.Additionally,SEAM was introduced to reinforce fea-ture fusion capabilities in the neck region.Additionally,a new loss function,Focaler-SIoU,was proposed to focus on samples of moderate difficulty,thereby improving the precision of bounding box regression.Valida-tion experiments conducted on a network dataset demonstrated that the improved model achieved a 10.3%in-crease in Precision(P),a 5.5%increase in Recall(R),and an 8.2%improvement in mean Average Preci-sion(mAP),significantly mitigating the issues of missed and false detections for occluded and small-scale tar-gets.关键词
深度学习/目标检测/YOLOv8/注意力机制Key words
Deep learning/Object detection/YOLOv8/Attention mechanism分类
信息技术与安全科学引用本文复制引用
文勇波,程凯伦,龙立阿,杨芳,范如谷,王玲..基于深度学习的门机运行工况人车检测算法研究[J].四川大学学报（自然科学版）,2025,62(2):369-376,8.基金项目
中国长江电力股份有限公司资助项目(5223020062) （5223020062）
