Image Recognition Research Of Intelligent Garbage Classification Based On Convolutional Neural Network

With the continuous development of social economy and science and technology,more and more garbage is produced in our daily life.It is of great significance for the reuse of recyclable resources and the protection of the natural environment to classify,store and treat these wastes.However,due to the lack of knowledge about garbage classification among most residents,the accuracy and efficiency of garbage classification at source are not high at present.In recent years,with the rapid development of convolutional neural networks,it is possible to realize the intelligent classification of garbage with deep learning technology.Garbage image has the characteristics of various categories and complex features,and the hardware and software resources of embedded devices are limited.In order to make the garbage image classification model more suitable for application in the classification trash,it is necessary to use the image classification model with small volume,which has poor self-training effect.Based on the above background,this thesis improved the garbage image classification model by adding the method of attention mechanism and knowledge distillation into the model.First,the garbage image classification model of ResNet+CBAM is proposed.On the basis of the structure of ResNet,the attention module of CBAM(Convolutional block attention module)is added.ResNet can efficiently transfer image features,and the attention module of CBAM can extract features from channel dimension and space dimension together.The accuracy of garbage image classification model reaches 91.23%,which is 4.19% higher than the original ResNet model on garbage image data set.Then for lightweight network MobileNet using knowledge distillation framework,through the "teacher" network-"student" network architecture,ResNet+CBAM model about garbage image knowledge distillation and transfer to MobileNet,greatly improve the MobileNet model on garbage image data set accuracy.The classification accuracy of MobileNet model is 93.53%,which is 2.3% higher than that of ResNet+CBAM.Meanwhile,the number of parameters in MobileNet model is only about one third of that of ResNet+CBAM model,and a garbage image classification model with lower number of parameters but higher accuracy is obtained.In this thesis,the two methods of adding attention mechanism and knowledge distillation are respectively used to improve the garbage image classification model,which effectively improves the accuracy of the model and obtains a garbage image classification model with high accuracy and small number of parameters.Therefore,this thesis provides a new idea for the improvement of garbage image classification model in theory,and also provides a further solution to solve the problem of garbage classification in practice.