| With the rapid development of information technology,information is full of people's daily life.The image occupies the most important position in all of the information,it contains a large amount of information,and easy to be accepte d by people.Compared to the voice and text,the image is more direct.Therefore,image processing has realistic significance and wide application prospect.In order to realize the task of image intelligent recognition,deep learning technology is used to solve the problem.Deep learning is one of the increasing attention researches in the field of machine learning,which simulates human visual processes and profoundly expresses the nature by inherent law and layer expression of sample data.In order to study the classification and description of some objects or process as the main content,DBN can directly use the original image recognition test,provide the basis for model recognition feature extraction and subsequent image to the original image,which is a kind of multilayer network structure.Based on deep theoretical research of DBN,this paper applies it to anomaly recognition of the fault indicator.In this paper,the expression of the weight and bias of DBN is derived from the energy formula,and the basic equivalent model of DBN is established.Meanwhile the expression explained the energy relationship between biases and signal,and the relationship between biases and probability.The limiting performance of classifica-tion were analyzed,and the method determining the number of hidden layer was given from the model of DBN.Results show that when the cross-correlation is 0(1),-1 or will never change of every two output hidden layer samples,there is no help to improve the classification accuracy by in creasing the number of layers further;Finally,the generalization of image is analyzed.The system test and a lot of tests are carried out by using the standard image and random image.The feasibility and practical value of the proposed method are verified by experiments. |
PDF Full Text Request