The Object Recognition Research Based on Convolution Neural Network

Abstract

Convolution neural networks include convolution computation and feedforward neural networks with depth structure. It is one of the most successful application fields of deep learning algorithm that can learn a lot of mapping relationship between input and output without any precise mathematical expression between input and output. The basic structure of convolutional neural network is input layer, convolution layer, pooling layer, full connection layer and output layer. Some improvements are proposed in this paper. First, a convolution layer and a pool layer are added to the original basic structure. Second, the new structure adopts hybrid pool in the pool stage. Thirdly, the maxout activation function is used in the full connection layer. Finally, based on the data set cifar-10, this paper studies the training and testing of convolutional neural networks for 10 categories of aircraft, horse, bird, ship, deer, dog, frog, automobile, cat and truck. The experimental results show that the accuracy rate of testing is 69.48%. Obviously the testing result is satisfactory.