Model And Analysis Of Garbage Image Classification Based On Machine Learning

In recent years,with the rapid development of China’s economic society,people’s living standards are gradually improved,while the economy is in booming,it is inevitable to begin facing a series of environmental governance issues.The pollution and influence caused by domestic garbage is increasingly serious,and the national government attaches great importance to this situation.Since 2019,as a pilot city,Shanghai has begun to take the lead in implementing garbage classification policies.At present,most of the garbage disposal methods are manual sorting,which cost manpower and long time.Due to this inefficient way,seeking more efficient and convenient garbage disposal programs have become important topics that are currently facing.With the rapid development of the current Internet,the intelligent technology classification of garbage image is gradually appeared.The development of machine learning algorithms and deep learning provide powerful technical support to image classification,and using algorithms to classify garbage images improve efficiency of waste treatment effectively.The garbage image dataset in this paper is from the public dataset of Kaggle and the network crawling,also the division of 80% is a training set,20% is a validation set.First,in the application of the traditional machine learning algorithm on the garbage image classification,this paper uses the SIFT algorithm to perform image feature extraction,also combined with SVM and XGBoost algorithm to construct garbage Image classification model,where in the two-categorous classification accuracy reaches: 78% and 85%respectively;the multi-class model is weak,reaching 55% and 50%;Multi-class model prediction accuracy is lower,reaching 55% and 50%;In view of the poor performance of traditional machine learning algorithms,we turn to deep learning technology.This paper uses convolutional neural network to establish an image classification dual output model,then the accuracy of multi-classification and second classification is 85%;Based on the EfficientNet-B0,the pre-training model of transformer learning is also established.This paper fine-tuned to the parameters of the model and rebuilt the fully connected layers.Through the training of the model,the dual output model reaches 82% accuracy,that is,transfer learning Methods have a certain impact on garbage images classification in accuracy and speed.Finally,based on the Vision Transformer,this paper has innovatively constructed the garbage image classification dual output model.The accuracy of the model reaches 77% and 90%,which fully verifies this algorithm has excellent effects on the garbage image classification.This paper’s research is the practice of traditional machine learning algorithms and deep learning in the garbage image classification.It is also an innovative attempt to take the front algorithm,which will provide a certain reference and support for the intelligent research of the garbage image classification.Selecting high-efficiency,predicted algorithm to build models can solve the automatic classification problem of daily garbage from the perspective of technology,in this way,the current garbage situation of spacious type,difficult to classify can become better,and it is also possible to further promote environmental protection and build a beautiful future in harmony between people and nature.