A Study On Waste Classification Model Based On Deep Convolutional Neural Network

As the economy develops,people’s standard of living increases,but at the same time it also causes a rapid rise in waste production,and the impact of waste on human health,the environment and economic development is self-evident.Relying on manual sorting is timeconsuming and ineffective,while the use of technologies related to deep learning makes it possible to automate waste disposal,which can effectively improve the efficiency of waste disposal through technological means.The key to waste disposal is to correctly identify and classify the waste,which not only improves the efficiency of comprehensive waste disposal,but also increases the value of waste resources for recycling while protecting the environment,so the task of classifying waste images is a meaningful study.In this paper,we use a lightweight convolutional neural network model to identify and classify rubbish after analysing the image classification model,which has too many network parameters,large computational effort,complex model and is not easy to port to mobile and embedded devices.This paper focuses on the following:（1）For the selection of the rubbish classification model,in view of the characteristics of the lightweight model with few parameters and fast speed,this paper selects Mobile Net V2 as the backbone network to classify the rubbish,based on which an Adam optimiser is introduced,a cross-entropy loss function is selected,and then the weights are updated using a backpropagation algorithm.This paper also establishes a rubbish dataset containing 245 subclasses and a total of 80,000 rubbish images,and normalises,greyscales,noise reduction and enhances them to improve the quality of the rubbish images.（2）In order to improve the accuracy of the model and enhance the learning of effective features while discarding useless features,this paper combines the attention mechanism with the trained Mobile Net V2 for training optimization.The feature maps outputted by the Mobile Net V2 network are fed into the CBAM module,so that the model can better learn useful and important features during the training process,and finally the obtained targets and features are analyzed using global average pooling and Softmax methods.Finally,a waste classification system is established to visualize the classification of the model.