Research on Garbage Image Classification and Recognition Method Based on Improved ResNet Network Model

Garbage classification recognition has great application value in daily life, while traditional image classification methods have low accuracy and generalization ability. This paper solves the above problems by improving the ResNet model. The specific strategy is to reduce its residual block and add it to the batch normalization layer (batch normalization), and then replace the convolution kernel in the first convolutional layer of the model with a convolution kernel. The ResNet34 (ResNet26) network model and the se attention mechanism module are added after each residual block, and finally an improved residual network model (AResNet26) is formed. On the dataset side, dataset augmentation is performed with the alumentations library. The results show that the AResNet26 model has higher recognition accuracy than ResNet26, ResNet18, ResNet34, VGG-16, AlexNet, and GoogleNet on the garbage image dataset. The recognition accuracy of the improved AResNet26 network model reaches 91.81%, and the training time is also greatly shortened. The research model realizes fast and accurate classification and identification of different types of garbage, and provides reference and technical support for garbage classification and identification methods.