This study aims to improve the accuracy of waste sorting through deep learning and to provide a possibility for intelligent waste classification based on computer vision/mobile phone terminals. A classification model of recyclable waste images based on deep learning is proposed in this paper. In this waste classification model, the self-monitoring module is added to the residual network model, which can integrate the relevant features of all channel graphs, compress the spatial dimension features, and have a global receptive field. But the number of channels is still kept unchanged; thereby, the model can improve the representation ability of the feature map and can automatically extract the features of different types of waste images. The proposed model was tested on the TrashNet dataset to classify recyclable waste and compare its classification performance with other algorithms. Experimental results show that the image classification accuracy of this model reaches 95.87%.

这项研究旨在通过深度学习提高废物分类的准确性，并为基于计算机视觉/移动电话终端的智能废物分类提供可能性。提出了一种基于深度学习的可回收垃圾图像分类模型。在此废物分类模型中，将自我监控模块添加到残差网络模型中，该模块可以集成所有通道图的相关特征，压缩空间维特征，并具有全局接受域。但是通道数仍保持不变；因此，该模型可以提高特征图的表示能力，并可以自动提取不同类型的废物图像的特征。在TrashNet数据集上对提出的模型进行了测试，以对可回收废物进行分类，并将其分类性能与其他算法进行比较。实验结果表明，该模型的图像分类精度达到95.87％。