Image Recognition Algorithm Based Ondeep Learning

Image is one of the most common information sources in daily life. Compared with other information sources, image contains much more amount of information which is complex and redundant. Processing image signal is usually difficult, but, however, human visual system has shown great ability in image processing. Many researcher has tried to build artificial neural networks for the simulation of the ability of human visual system by bionic methods. After the success of shallow artificial neural network, the study of deep artificial neural network has encountered some difficulties such as high cost of training or local optimal solutions. New advance in deep neural network come with deep learning models, which have brought researchers new inspirations and enthusiasm.Based on the study of deep belief networks (DBN), this paper raised a new cross-field deep belief network model, which could recognize images from multiple problem fields. According to the truth that DBN captures image features level-by-level, this paper used a special method to combine DBNs of multiple fields. The bottom level of DBNs are shared to capture the low-level features in data space, while the high level neurons are still detached to capture the high-level features of the specific fields. Besides capturing low-level features, shared neurons also lead input data into correct high level neurons, which will finally make the reorganization in specific fields. This cross-field DBN is trying to simulate human visual system, which has bionic meanings. Meanwhile, cross-field unlabeled sample sets could be used to train this model due to the shared neurons, which could improve the recognition performance in current internet environment with large scale cross-field unlabeled images. Because of the detached neurons, paralleled computing is possible, which could reduce the time cost of training.This paper designed and realized the cross-field DBN with Theano library, and tested its performance in MNIST data set of handwriting characters and COIL data set of object images. Experiment proofed that the cross-field DBN has the same performance with the traditional DBN. But, with the help of cross-field unlabeled training data, cross-field DBN has better performance than traditional ones. This paper also figure out that in a deep model, the supervised training of bottom neurons could be replaced by unsupervised training. According to this, some improvements are applied on DBN training algorithm, which made the paralleled computing of cross-field DBN easier. This paper realized cross-field DBN model’s parallelization by Hadoop framework. Experiment proofed that, under the condition that the scales of each field’ s sample set are roughly equal, parallelization could reduce time cost of training effectively.