Research On Garbage Classification Based On Convolutional Neural Network

In recent years,the Chinese government has attached great importance to the construction of ecologically civilized cities,and the classification of domestic garbage is an important part of it.The survey shows that most residents have the awareness of domestic garbage classification,but due to the lack of relevant domestic garbage classification knowledge and guidance,domestic garbage cannot be effectively treated,which has an important impact on the urban ecological environment.The development of deep learning has provided a brand new idea for the classification of domestic garbage.Image classification is an important and challenging research topic in the current deep learning research field,and it also has a wide range of application prospects.This paper is based on the convolutional neural network to research the classification of domestic garbage,and combined with the research results to design a domestic garbage classification small program to assist users to efficiently complete the classification of domestic garbage.The main research work is as follows:(1)Aiming at the problems of the current convolutional neural network model in the image classification field with too many model parameters and long computing time,this paper proposes a garbage classification network model(Garbage Net-V2)based on deepthwise separable convolution.This model uses deepthwise separable convolution to replace the standard convolution of some layers in the Res Ne Xt-101 network model,which effectively reduces the parameters and calculations of the network model.At the same time,the model also reformed the high-level structure of the Res Ne Xt-101 network,replaced the fully connected layer with standard convolution,and mapped the final output channel to the garbage category,reducing the parameters and giving each channel practical meaning.Finally,through experimental verification,the accuracy of Garbage Net-V2 network and Res Ne Xt-101 network are basically the same,but the size of Garbage Net-V2 has been significantly reduced.(2)The feature representation difference between the same category is not obvious,and there is feature redundancy,which is not conducive to the accurate classification of pictures.This paper proposes a multi-scale channel attention mechanism(MSE-Net)that improves the SENet module.While paying attention to the contribution of different channels to image classification tasks,MSE-Net introduces the concept of multi-scale.It is believed that the contributions of different scales of different channels to image classification tasks are also different,so as to obtain more detailed information.Finally,MSE-Net is embedded in some mainstream image classification networks and the Garbage Net-V2 network in this article.The experimental result analysis of the model embedded in MSE-Net verifies that the MSENet module has good results.(3)Aiming at the landing problem of research results,a garbage classification small program based on the research results of this article is designed and implemented.The garbage classification small program is composed of data collection and sending module,garbage classification module,and database module.After functional testing of each module,the garbage classification small program implemented in this paper can identify domestic garbage online in real time,stably and accurately,and has certain application and promotion value.