Research On Image Recognition Algorithm Based On Sparse Representation And Deep Learning

With the rapid development of computer technology,physics and sensor technology,storage technology,and network technology,the expansion of data has become increasingly intensified.There are massive image(video)data generated from time to time when mobile phones are used for surveillance cameras of various purposes and in various occasions.The research of smart mobile robotsis also constantly being collected.And accumulated experimental and test images and videos.From the perspective of signal processing,images are high-dimensional data and carry a large number of complex information and features.How to efficiently analyze and identify image data and make reasonable use of it is a problem that needs urgent study and solution.Sparse representation is a novel signal processing technology proposed in recent years,and compressive sensing lays a theoretical foundation for engineering application.At present,sparse representation has been widely used in computer vision,machine learning and other fields,and has achieved beneficial results.For example:image compression,image denoising,image super-resolution,image reconstruction,image classification and recognition.In this paper,an in-depth study of image recognition algorithms and applications based on sparse representation and deep learning is carried out.The main work is summarized as follows:1.Image sparse representation and deep learning methods.Summarized the basic framework and method of image sparse representation;focused on research and summary of the compressive sensing methods,including signal sparse representation,design observation matrix,signal reconstruction theory.At the same time,the basic functions of deep learning are described and summarized.2.A sparse representation classification recognition algorithm based on Log-Gabor is proposed.By integrating the global features of the sample data and the local feature information,the recognition rate can be improved.In the framework of compressive sensing and sparse representation,the algorithm uses the linear correlation between the same class of samples in its characteristic subspace,solves the global representation of the sample to be identified on all training samples,and constructs a dictionary using global features.Combined with the local features of the sample,the sparse decomposition and expression of the identified samples are used to make the sparse representation vectors as specific as possible,able to cope with the situation that each type of training sample is small,reduce the corresponding number of categories,and improve the classification and recognition results,accuracy.Experiments in face recognition and traffic sign recognition show that the Log-GSRC algorithm significantly improves the recognition rate over the SRC algorithm as the number of training samples and the number of sample classes increase.3.A sparse representation classification recognition algorithm based on dictionary learning is proposed.An effective sparse representation super-complete redundant dictionary is designed by dictionary learning algorithm.It can deal with the situation that most of the sample data is relatively small and solve the problem to some extent.The problem of large image variability caused by environmental changes,resulting in a sparser and more accurate representation of the image to be recognized,and then using a sparse representation classifier for image recognition.The algorithm is applied to face recognition and traffic sign recognition.Experimental results show that the recognition rate increases significantly with the number of training samples as the number of training samples increases.4.A sparse representation classification recognition algorithm based on deep learning features is proposed.The sample features are extracted for large sample data and network training is performed.A deep convolutional network model is designed and sparse expression methods are used to effectively achieve a certain degree of effectiveness.Remove the effect of unstable image quality caused by some uncertain factors such as illumination,obstruction,demeanor and posture.The algorithm uses the deep learning method to extract the sample feature information,and obtains an optimized deep convolutional network model through sample training.Then,the sparse representation algorithm is used to classify and identify the test set.Experiments show that the algorithm has a significant improvement in recognition rate in small samples and complex environmental changes.