安检图像小目标违禁品特征提取模块构建与应用  

Construction and Application of a Feature Extraction Module for Small Target Prohibited Items in Security Inspection Images

作　　者：刘天时[1] 周泽华 郝敏杰 LIU Tianshi;ZHOU Zehua;HAO Minjie(School of Computer Science,Xi'an Shiyou University,Xi'an 710065,China)

机构地区：[1]西安石油大学计算机学院,陕西西安710065

出　　处：《现代信息科技》2024年第4期136-141,共6页Modern Information Technology

摘　　要：针对物流包裹安检图像中小目标违禁品易漏检问题,通过在感受野模块的多分支并行网络上引入卷积注意力模块,构建一种适用于小目标违禁品检测的特征提取模块。在此基础上,将构建的特征提取模块融入YOLOv5模型的主干部分,使得模型在违禁品检测的过程中聚焦于图像的重要特征。为了充分发挥所构建模块对于小目标物体的特征提取能力,采用空间深度转换模块替代原模型中的下采样模块,使得YOLOv5模型在特征提取的过程中能够尽可能地保留小目标物体的特征信息,提高对小目标违禁品的检测效果。Aiming at the problem that small target prohibited items in logistics package security inspection images is easy to miss detection,a feature extraction module suitable for small target prohibited items detection is constructed by introducing a convolutional attention module on the multi-branch parallel network of the receptive field module.On this basis,the constructed feature extraction module is integrated into the backbone of the YOLOv5 model,so that the model focuses on the important features of the image in the process of prohibited items detection.In order to fully utilize the feature extraction ability of the constructed module for small target objects,a spatial depth conversion module is used to replace the downsampling module in the original model,so that the YOLOv5 model can retain the feature information of small target objects as much as possible during the feature extraction process,and improve the detection effect for small target prohibited items.

关 键 词：安检图像 小目标违禁品 特征提取模块 计算机视觉 物流包裹 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]