重庆邮电大学学报（自然科学版）2025，Vol.37Issue(1)：46-56,11.DOI:10.3979/j.issn.1673-825X.202312180422
基于WHIoU损失的安检图像违禁物品检测算法
WHIoU loss-based detection algorithm of prohibited items in security check images
摘要
Abstract
To address the issues of overlapping prohibited items and complex backgrounds in X-ray security inspection ima-ges,this paper proposes a modified X-ray inspection image detection algorithm based on YOLOv7.An attention feature fu-sion module is constructed in the neck network of the model,where both spatial and channel attention mechanisms are ap-plied to extract shallow detail features and deep semantic features,reducing noise information redundancy and minimizing the loss of effective features.Additionally,the WHIoU Loss is redesigned to replace CIoU Loss as the new bounding box loss function.This modification ensures that the aspect ratio penalty function still constrains the model when the predicted and ground truth boxes maintain the same linear proportion in width and height,improving convergence speed and accuracy.Experimental results on the SIXray dataset show that the algorithm achieves a 2% improvement in mAP50 and a 4.5% im-provement in mAP50:95,while maintaining a detection speed of 58 FPS.关键词
深度学习/目标检测/YOLOv7/安检图像/注意力机制Key words
deep learning/object detection/YOLOv7/security check image/attention mechanism分类
信息技术与安全科学引用本文复制引用
朱雷,李国权,武瑞恒,黄正文,庞宇..基于WHIoU损失的安检图像违禁物品检测算法[J].重庆邮电大学学报（自然科学版）,2025,37(1):46-56,11.基金项目
国家自然科学基金项目(U21A20447) （U21A20447）
重庆市创新群体科学基金项目(cstc2020jcyj-cxttX0002)Natural Science Foundation of China(U21A20447) （cstc2020jcyj-cxttX0002）
Foundation for Innovative Research Groups of Natural Science Foundation of Chongqing(cstc2020jcyj-cxttX0002) （cstc2020jcyj-cxttX0002）
