Research On Algorithm And Application Of Deep Learning Based On Convolutional Neural Network

Deep learning is a new research direction in the field of computer science and machine learning. It was introduced into machine learning, and makes machine learning closer to its original goals:artificial intelligence. Deep learning learns inherent rules and the abstract hierarchies of sample data. The hierarchies can be used to help explain the data, such as text, images and sounds. It aims to let the machine acquire learning ability to cognize text, image and voice data, etc.As a complex machine learning algorithms, recognition accuracy of deep learning on image and audio is far beyond the previous technologies. Deep learning has also achieved great success in the search technology, data mining, machine learning, machine translation, natural language processing, multimedia learning, speech, recommendation and personalization technology, etc. It has solved many complicated pattern recognition problem, and promoted the progress of the artificial intelligence technology. It is a very important work to combine deep learning with application research.The development of deep learning and the latest research results at home and abroad have been reviewed and summarized in this article. The concept and algorithm of artificial neural network and classical convolutional neural network was briefly introduced. The convolutional neural network algorithm is improved and applied in OCR (Optical Character Recognition) and TSR (Traffic Sign Recognition). The structure and performance of deep neural network are studied respectively from theory and application. The main work in this paper is as follows:1. The neural network is improved based on LeNet-5network model in this paper. A number of convolutional neural network models are constructed, each with different style of layer connection and different number of neurons in the convolutionl layer for feature extraction. Each model is applied to the issue of optical digital recognition, and then we analyze and compare performance of various models through learning process and the efficiency of recognition in the experiment.2. Based on the ideology of adaptive enhancement (Adaboost), a multi-column convolution neural network model is build. It is used in the traffic sign recognition. The preprocessed data is used to train the multi-column convolutional neural network to realize high performance of traffic sign recognition. 3. Finally, we validate application feasibility of convolution neural network on optical character recognition and traffic sign recognition through the experiment. It is compared with other state-of-the-art recognition algorithms. The efficiency of the convolutionl neural network is analyzed in the actual engineering application.