基于机器学习的垃圾分类

Garbage classification based on machine learningAbstract:In 2019,the garbage classification started from Shanghai and then implemented nationwide,mainly to classify the types of garbage.The classification includes recyclable,food waste,hazardousand others.This article takes garbage classification as the core to carry out a series of investigations,using machine leaming to classify garbage,mainly to identify white pollutants and harmful garbage.Using convolutional neural networks and other functions of machine learning,with the help of deepleaming and image recognition,the garbage is made into a series of research and processing,and finallyJupyter shows which type of garbage this picture is.In this paper,the Xception model is used to identify harmful garbage and white pollutants byconvolutional neural networks.Xception only supports TensorFlow as a backend.Therefore,Jupyter isused for display.Xception is mainly to eliminate the correlation between the cross-channel correlationexisting in the feature maps in the convolutional neural network and the spatial correlation.It is a veryeasy to use convolutional neural network,which has a very significant effect on improving therecognition effect.Use a series of auxiliary functions,such as auxiliary functions for adding directoryand file names,auxiliary functions for drawing,etc.Use scikit-leam to calculate weights to balance thedata set as much as possible.The low-level convolutional neural network can recognize different shapesin the image.It is the classification of the last few fully connected layers that combine these featuresinto the entire image.Therefore,the output of the last convolutional layer of the model is reconnected toa new fully connected neural network and used to classify the data set.Keras records performanceindicators at the end of each "epoch"for later plotting.During training,the loss value of the training setis usually reduced,but the loss value of the test set is a bit unstable.The same is that,on the training set,the classification accuracy is usually improved,while the test set is somewhat unstable.After training,asingle function in the Keras API is used to evaluate the performance of the new model in the test setKeywords:Garbage classification,convolution neural network,image recognition,image processing,Xception model,transfer leaming