Reseacrh On Image Recognition Based On Deep Learning Algorithm

Image recognition technology has been widely applied and played animportant role in various fields nowadays. Good recognition technology is thekey. To improve the recognition rate and speed have great significance, whichinfluence image recognition practicality and safety directly.Because of multi-layer structure of deep network can use a more conciseway to express complex functions. Therefore, deep learning will be applied tothe image recognition to improve the accuracy of image recognition. Firstly,support vector machine is combined with deep learning to build the model forimage recognition. Then, use the convolutional restrictive Boltzmann machine tobuild deep network, and improving the training process.The main work of this paper as follows:(1) Analysis the image recognition methods and existing problems carefully.Existing image recognition algorithms were compared. Introduce thedevelopment and progression about depth study. The deep learning structurewith the shallow structure is compared. Common method of deep learning issummarized. Introduced the principles and training process of restrictedBoltzmann machine and convolution restricted Boltzmann machine. (2) Combined support vector machine with deep learning to constructmultilayer deep networks for image recognition. Through experiments, theresult is well compared with support vector machines and deep belief networks.Compared different number of samples, layers, nodes with the correct rate andthe influence of SVM parameters. Clarify the relationship of hidden layer nodeswith numbers of support vector.(3) The convolution restricted Boltzmann machine is used to build the deepnetwork. Then unsupervised and supervised learning alternately to train thenetwork to improve the training process. And this network applied into imagerecognition. Through experiments verify the feasibility and effectiveness.