Image Spam Classification Based On Deep Learning

With the accelerated development of science and technology,network information experience explosion,spam problem is increasingly serious,more and more attention,traditional spam classification and recognition technology mainly for text spam.However,in recent years,spammers have added a lot of spam information to their images and have been processed through some blurring techniques to escape the traditional text-like spam classification and recognition technology,called image-type spam.This paper is devoted to the classification of image spam,hereinafter referred to as junk images,such as:(1)Research on convolutional neural network optimization algorithm: Combined with theoretical knowledge and experimental research,a set of convolutinal neural network optimization scheme.Firstly,the generalization ability of the network is affected by the diversity and size of the training data set.During the network training process,it can be augmented according to the actual situation to improve the generalization ability of the network model;Then,the expression ability of the network is affected by its scale and complexity,and it can solve the more complex classification problems by improving the neuron scale and layer depth of the network;Finally,the over fitting problem of complex models during training is solved using the regularization layer and data distribution optimization.(2)Features of SIFT-CNN garbage image extraction: According to the actual situation of garbage image data set,its network structure and algorithm are optimized based on Alex Net.However,the complex network structure has the problem of high computational complexity.In contrast,although traditional SIFT features have limited ability to represent images,their computational complexity is relatively low.In this paper,we combine shallow features with deep learning features to extract the SIFT-CNN features of garbage images.(3)The final implementation and performance analysis of the classification model: After the SIFT-CNN features of garbage images are extracted,the support vector machine is used as the final classifier of the model,and then its parameters are optimized.Finally,the experimental model,extraction rate and recall rate were compared and analyzed.To a certain extent,the accuracy of detection is improved,and the false detection rate is significantly reduced.Therefore,the classification model studied in this paper has better classification effect and performance.