多特征加权融合算法的生活垃圾图像分类研究  

Garbage Image Classification Based on Multi-feature Weighted Fusion Algorithm

作　　者：王敏[1] WANG Min(Fujian Vocational College of Shipping and Transportation,Fuzhou,Fujian 350007,China)

机构地区：[1]福建船政交通职业学院,福建福州350007

出　　处：《河北北方学院学报（自然科学版）》2023年第3期27-31,36,共6页Journal of Hebei North University:Natural Science Edition

摘　　要：为改善我国现有的垃圾问题,实现准确且高效的垃圾分类工作,提出一种基于多特征加权融合的生活垃圾图像分类算法。该算法利用ResNet网络为主干网络,利用多分支网络结构提取图像不同层次的特征信息再进行加权融合,并进行了自适应权重融合实验和固定权重融合实验。实验结果表明,自适应实验组的分类准确率为97.29%;进行固定权重融合实验,发现全部平均系数下模型的生活垃圾图像分类准确率为97.42%;对比不同算法模型下分类的准确率大小,得到此次研究提出模型的分类准确率为97.52%。说明所提出的算法模型能够较好地识别各种生活垃圾图像,并对垃圾图像进行高效分类。To solve the existing garbage problem in China and achieve accurate and efficient garbage classification,a garbage image classification algorithm based on multi-feature weighted fusion is proposed.The algorithm used ResNet network as the backbone network and used multi-branch network structure to extract different levels of image feature information.And then it carried out weighted fusion and adaptive weight fusion experiments and fixed weight fusion experiments.The experimental results showed that the classification accuracy of the adaptive experimental group was 97.29%;Fixed weight fusion experiment showed that the classification accuracy of the model under all average coefficients was 97.42%;Compared with the classification accuracy of different algorithm models,the classification accuracy of the model proposed in this study was 97.52%.It showed that the algorithm model proposed in this study could better identify various garbage images and classify garbage images efficiently.

关 键 词：多特征 加权融合 图像分类 深度学习 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]