Machine Learning Based Recyclable Garbage Identification Classification Research

With the development of modernization,the increasing amount of household garbage has seriously polluted the environment,which is a serious threat to human health and restricts the development of economy.Scientific classification of recyclable garbage can not only reduce environmental pollution,improve the quality of life but also save resources.At present,garbage sorting relies mainly on manual sorting,which has many disadvantages such as low efficiency and high misjudgment rate.By using machine learning algorithm,these problems are effectively solved,which brings convenience to people.In this paper,classical feature extraction method and deep learning algorithms are used to discuss the problem,combined with the created dataset,a new recyclable garbage classification model is designed,which has been proved to improve the accuracy of garbage classification.The research content of this paper mainly includes:(1)Constructing a garbage classification dataset.Concerning the imbalance,single background and overly obvious features of the existing garbage classification dataset,the garbage dataset REGB is created based on the classical dataset Trash Net category,and a series of data enhancement strategies are carried out to provide the data basis for the classification model.(2)The traditional manual extraction method is used to extract the HOG operator and LBP operator vectors of the garbage images and classify them with the support vector machine algorithm,the highest classification accuracy of the two extraction methods are 64.2% and 62.2% respectively.In order to improve the classification accuracy of the model,the two extracted features are fused using the tandem method and the PCA is used to reduce the dimension.As a result,the accuracy is improved by0.9%,but the overall effect still needs to be improved by deep learning algorithm.(3)A new network model A-Res Net is proposed.firstly,the deep learning framework Py Torch is selected to study VGG16,Mobile Net,Xception and Res Net50,the transfer learning is adopted to reconstruct the the fully connected layer and finetuned the parameters of each model,after comparing the performance of each model,Res Net50 with better classification effect is selected as the baseline model.Then aiming at the problem that feature differentiation between different categories in the recyclable garbage images is inconspicuous,combining with transfer learning,the attention mechanism module CBAM is innovatively added to the structure of the residual network Res Net50 at several different positions and Dropout layer is added,which finally achieve the effect of reinforcing important channel information and prevent overfitting.Experiments show that the A-Res Net model can effectively improve the classification accuracy of recyclable garbage,with a final classification accuracy of 96.7%,achieving the expected results.