Research On Classification Of Garbage Image Based On Deep Learning

With the rapid development of China’s economy and the continuous acceleration of the urbanization process,the output of urban household garbage is increasing year after year,and it has become the country producing the most garbage in the world.For a long time,the awareness of garbage classification is weak,which makes the garbage siege more and more intense.The classification of municipal solid waste is the basic requirement of the transformation from waste disposal to resource utilization,and it is also an effective way to promote environmental governance.In the past few years,there were still a large number of scavengers in China,who were responsible for some of the classification work of urban household garbage.However,in recent years,overcapacity and the decline of raw material prices have directly affected the price of waste products,leading to the gradual withdrawal of scavengers from our lives,which has objectively made garbage classification more urgent.To promote the smooth development of urban garbage classification,on the one hand,we need to continuously improve the system and the level of awareness,on the other hand,we also need to improve the ability of accurate classification of household garbage.However,garbage classification only relies on residents’ existing cognition,which inevitably leads to the wrong classification of garbage.With the progress of science and technology,it is no wonder that the work of image classification is done by machines.Most of the traditional machine learning methods for image classification are based on the classification of image features,that is,according to the differences of different types of images,the image processing algorithm is used to propose the corresponding qualitative or quantitative features,and the classification of these features.The quality of the image classification is largely determined by the quality of the extracted features,and the extracted image features will vary greatly according to the different experimenters.The accuracy and stability of image classification can be greatly improved by automatic feature extraction.In recent years,deep learning technology has made very rapid development.Deep learning mainly builds neural networks by imitating the process of connection and communication between biological neurons,so as to realize the analysis and processing of data.These network layers can realize automatic extraction of image features,so deep learning has the advantage of higher accuracy than machine learning in image classification.Deep learning technology is playing an increasingly important role in the field of image classification.Therefore,this paper adopts the method of deep learning to classify household garbage images.First of all,this paper makes an exploratory data analysis on the image data set of domestic garbage,mainly for the statistical analysis and visual display of the distribution of image categories and image sizes:the domestic garbage images in this paper are classified and predicted according to 40 sub categories of other garbage,kitchen waste,recyclable and harmful garbage,which belong to the multi classification and prediction question of 40 classification The width/height ratio of the data is about 1.Then the data is enhanced pertinently.Firstly,the image is zoomed,then the image is flipped horizontally and vertically with a probability of 50%.Finally,the image is rotated with a random probability.After image enhancement,AlexNet,VGG,ResNet50 and ResNeXt101 were first used for model training,and through the comparison experiment with or without transfer learning,it was concluded that the accuracy of the model after transfer learning was higher than that of the model without transfer learning,and ResNeXt101 had the highest accuracy among the four convolutional neural networks.Then,the ResNeXt network architecture was improved by adding an attention module for each residual in the network,so that the model could focus on extracting image information.The experiment proves that the improved ResNeXt has a higher accuracy in the classification of household garbage images.Finally,the Web framework of garbage image classification constructed by Flask was sent to postman,and the online prediction of MSW image classification was realized by inputting garbage image.