基于双分支网络的户外垃圾检测识别  

Outdoor Garbage Detection and Recognition Based on Dual Branch Networks

作　　者：赵文琪 张立新 阚希 郑好韧 ZHAO Wenqi;ZHANG Lixin;KAN Xi;ZHENG Haoren(School of Automation,Nanjing University of Information Science and Technology,Nanjing 210044,China;School of IoT Engineering,Wuxi University,Wuxi 214063,China)

机构地区：[1]南京信息工程大学自动化学院,江苏南京210044 [2]无锡学院物联网工程学院,江苏无锡214063

出　　处：《电子科技》2025年第4期1-9,24,共10页Electronic Science and Technology

基　　金：国家自然科学基金(42105143);江苏省高等学校基础科学研究面上项目(580221016)。

摘　　要：现有户外垃圾检测算法在特征提取方面没有充分考虑CNN(Convolutional Neural Network)和Transformer的优缺点,限制了网络的整体性能。文中提出了一种由CNN和Transformer组成的双分支融合网络检测算法。在编码阶段,基于CNN和Transformer的优点构建了双分支主干网络来对原始图像的特征信息进行提取。通过多尺度卷积模块和多尺度池化模块消除所提取特征信息在维度和语义上的差异,通过加强特征提取网络减少深层神经网络中细节信息的丢失。文中收集了6个类别的户外常见垃圾图片,建立了一个背景复杂的垃圾图片数据集来验证所提算法在户外垃圾检测识别任务上的性能。实验结果表明,相比最新目标检测算法,所提算法在该数据集上的平均准确率(mean Average Precision,mAP)提升约5%。为了验证所提算法的泛化性能,在华为垃圾分类挑战杯数据集上进行了泛化实验,实验结果表明,相比最新目标检测算法,所提算法的mAP提升了约2%。The existing outdoor waste detection algorithms do not fully consider the advantages and disadvantages of CNN(Convolutional Neural Network)and Transformer in feature extraction,which limits the overall performance of the network.This study proposes a two-branch fusion network detection algorithm composed of CNN and Transformer.In the coding stage,a two-branch backbone network is constructed based on the advantages of CNN and Transformer to extract the feature information of the original image.Multi-scale convolutional module and multi-scale pooling module are used to eliminate the differences in dimension and semantics of extracted feature information,and the loss of detail information in deep neural network is reduced by strengthening feature extraction network.Six types of outdoor garbage images are collected,and a data set of garbage images with complex background is built to verify the performance of the proposed algorithm in outdoor garbage detection and recognition task.The experimental results show that the mAP(mean Average Precision)of the proposed algorithm on this data set is improved by about 5%when compared with the latest target detection algorithm.In order to verify the generalization performance of the proposed algorithm,a generalization experiment is carried out on the Huawei garbage classification challenge cup data set,and the experimental results show that mAP of the proposed algorithm is improved by about 2%when compared with the latest object detection algorithm.

关 键 词：垃圾污染 自动检测识别 CNN TRANSFORMER 双分支融合网络 多尺度卷积 多尺度池化 特征信息 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] X705[自动化与计算机技术—计算机科学与技术]