Application of alternating deep belief network in image classification

Aiming at the problem that the bottom layer parameters of deep belief network (DBN) can not be fully learned during the process of image classification, this paper proposes an image classification method based on alternating deep belief network (ADBN). After the unsupervised learning of each layer of restricted boltzmann machine (RBM), we use back propagation (BP) algorithm to fine tuning its parameters. Alternating the use of unsupervised and supervised training process, so that the entire network weights can achieve minimum training error. Consequently, a comparative experiment is conducted on multiple data sets of UCI database. Experimental results indicate that ADBN has effectively alleviated the problem of vanishing gradient and obtained higher accuracy than DBN and support vector machine (SVM).