Compressed Sensing-based Classification

Image classification algorithms are the core branch of the machine learning field.Compared with data mining and natural language processing,the application of image algorithms is more extensive.However,for a large amount of image data,achieving a complex classification model still requires a very high computational power and time cost.Therefore,based on the theoretical basis of compressed sensing,this paper proposes an image classification algorithm with greatly improved computational speed while ensuring the accuracy of model classification.Meanwhile,this paper designs and implements an image classification system covering modules such as image acquisition,preprocessing,compression,training and pre-diction.The image classification system was applied to the gait recognition and multi-resolution image classification which achieved good results.The research work of this paper is as follows:1)Theoretical research on compressed sensing:Explain various methods for Sparse expres-sion of signal,selection of sensing matrix and image reconstruction.Simulated the above methods to achieve image undersampling and reconstruction;2)Combining compressed sensing with image classification,and an image classification method based on compressed sensing is proposed.Since the image classification does not require compressed sensing to completely reconstruct the original image,as long as the classifica-tion is completed,the compressed data can be directly classified in the compressed domain of the compressed sensing,thereby greatly reducing the amount of processed data.For the accuracy rate of this algorithm,this paper utilized the requirements of Restricted Isometry Property in compressed sensing to derive Isometricity of data discrimination between im-ages in original space and projection space.This means that in the compressed domain,the discrimination between different samples is unchanged,and the accuracy of image classi-fication is not affected.In addition,the traditional image classification algorithm and the comparative simulation experiment of this algorithm show that the classification accuracy of the two is the same,while the latter takes much less time than the former;3)The image recognition algorithm based on compressed sensing is applied to the actual hu-man gait recognition system.The system is dedicated to collecting pedestrian gait in the environment of corridors,offices,prisons,etc.,and the system automatically determines the identity of the person.The algorithm greatly reduces the time of data update and model training of the system,and realizes real-time discrimination.4)Further,in this algorithm,since number of dimensions of the compressed image is control-lable,a multi-resolution image classification preprocessing method based on compressed sensing is proposed.Comparing it with the traditional image dimensionality reduction meth-ods,the accuracy of the former in the multi-resolution image classification model is much higher than the traditional methods.