电子科技2025，Vol.38Issue(11)：42-52,11.DOI:10.16180/j.cnki.issn1007-7820.2025.11.006
基于改进YOLOv7的安检小目标检测方法
A Method for Detecting Small Objects in Security Inspection Based on Improved YOLOv7
摘要
Abstract
In view of the problem of low detection accuracy of small target objects in security checks,this study proposes a security check small target detection model XS-YOLO(Xray-Spection-YOLO)based on improved YOLOv7(You Only Look Once version7).The feature extraction backbone network SwinT(Swin Transformer Backbone)based on self-attentional global perception is used to construct global attention and obtain feature information.FHD(Four Head Detection)based on multi-scale hierarchical feature fusion is proposed.A fourth detection head is introduced to improve the model's ability to identify small targets.A branch detection strategy DCSP(Decoupled Cross Stage Partial)based on the decoupling of classification and location tasks is proposed,and the output branches after feature fusion are processed by feature maps respectively.The experimental results show that compared with YOLOv7,the mAP(mean Average Precision)of XS-YOLO on HiXray data set has increased by 2.8 percentage points,and the mAP of XS-YO-LO on OPIXray data set has increased by 5.8 percentage points.关键词
深度学习/计算机视觉/目标检测/YOLOv7/安检/危险物品检测/小目标检测/自注意力机制Key words
deep learning/computer vision/target detection/YOLOv7/security check/detection of dangerous goods/small target detection/self-attention mechanism分类
计算机与自动化引用本文复制引用
杨钧彦,徐向华..基于改进YOLOv7的安检小目标检测方法[J].电子科技,2025,38(11):42-52,11.基金项目
浙江省尖兵领雁计划(2022C03132) Zhejiang Province Leading Wild Goose Plan(2022C03132) （2022C03132）
