Applying Improved Convolutional Neural Network in Image Classification

In order to solve the poor accuracy problem which caused by the gradient descent easily fail into local optimum during the training process and the noise interference in process of feature extracting. This paper presents an integrated optimization method of simulated annealing (SA) and Gaussian convolution based on Convolutional Neural Network (CNN). Firstly, the improved algorithm extract some features from the central feature of a model as priori information, and find the optimal solution as initial weights of full-connection layer by simulating annealing, so as to accelerate the weight updating and convergence rate. Secondly, using the Gaussian convolution method, this paper can smooth image to reduce noise disturbing. Finally, the improved integrated optimization method is applied to the MNIST and CIFAR-10 databases. Simulation results show that the accuracy rate of the integrated network is improved through the contrastive analysis of different algorithms.