基于改进深度学习模型的城市建设管理中废弃物精细分类智能框架  

An Intelligent Fine Garbage Classification Framework for Urban Construction Management Based on Improved Deep Learning Models

作　　者：周春媚 ZHOU Chunmei(Faculty of Public Administration,Guangxi Police College,Nanning 530299,China)

机构地区：[1]广西警察学院公共管理学院,南宁530299

出　　处：《北京印刷学院学报》2024年第8期55-60,共6页Journal of Beijing Institute of Graphic Communication

基　　金：2024年度广西高校中青年教师科研基础能力提升项目“基于深度学习的多视角三维重建算法研究”(2024KY1207)成果。

摘　　要：针对城市建设管理中的自动化废弃物分类,本文提出基于改进Swin Transformer(SWT)和卷积神经网络(CNN)的深度学习框架,利用ST作为废弃物分类的基准模型,利用层级Transformer结构,通过平移窗口计算特征表征。由此,在保持局部性的同时,允许跨窗口连接,提高图像处理效率。在每个ST模块的末端添加特征强化(FS)模块,提高特征提取能力,并通过加权交叉熵损失函数的设计解决类别不平衡问题。公开废弃物分类数据集上的性能结果表明,所提框架在准确度、精度和召回率指标上取得了最好性能,分类准确率达到了98.93%,可实现有效的自动化废弃物分类,提高废弃物处理效率,提升城市形象,推动城市向着更加智慧、绿色和可持续的方向发展。For automated garbage classification in urban construction management,a deep learning framework based on improved Swin Transformer(SWT)and Convolutional Neural Network(CNN)is proposed.The SWT is utilized as the benchmark model for garbage classification,leveraging hierarchical Transformer structure to compute feature representations through shifting windows.Thus,while maintaining locality,cross-window connections are allowed to improve image processing efficiency.A Feature Strengthening(FS)module is added at the end of each SWT module to enhance feature extraction capabilities.The problem of class imbalance is addressed through the design of a weighted cross-entropy loss function.Performance results on a publicly available garbage separation dataset show that the proposed framework achieves the best performance in terms of accuracy,precision and recall,with a classification accuracy of 98.93%,surpassing current state-of-the-art methods.The proposed method enables effective automated garbage classification,improves waste processing efficiency,enhances urban image,and promotes the development of cities towards a smarter,greener and more sustainable direction.

关 键 词：深度学习 Swin Transformer 卷积神经网络 废弃物分类 特征强化 平移窗口 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]