Deep Neural Networks for Garbage Classification

Effective garbage classification and recycling are of great significance for protecting the environment, conserving resources and improving people's livelihoods, and have a far-reaching impact on promoting sustainable development. However, the traditional waste classification method has problems such as low accuracy and low efficiency, while its reliance on manual labor and high time consumption make it difficult to adapt to the needs of modern society for waste classification. To address this problem, we propose in this study to utilize deep convolutional neural networks, mainly Residual Network with 18 layers (ResNet18) model to improve the efficiency and accuracy of garbage classification. In this study, ResNet18 is first evaluated comprehensively for several key performance metrics in the waste classification task, including: Accuracy (Accuracy, Recall, Precision, F1-score, Kappa coefficient, Params and FLOPs. Then, we also evaluate the performance of classical convolutional neural network models such as LeNet, Visual Geometry Group Network with 19 layers (VGG19), AlexNet, and MobileNet, and analyze them in comparison with ResNet18, which shows that ResNet18 has a superior performance in dealing with the waste classification The results show that ResNet18 has superior performance in the task of garbage classification. Finally, we analyze the reasons why ResNet18 exhibits the best performance and propose directions for further optimizing the model performance, which provide directions for future research and experiments.