An image recognition method using parallel deep CNN
Zhiqiang Gao, Yuexin Li, Shijie Wang · Academic Journal of Computing & Information Science · 2019
Based on the characteristics of the image, the traditional methods are improved based on the theory of convolutional networks and the research results at home and abroad. Firstly, the training algorithm of convolutional network is studied. By analysing the algorithm, debugging and finding the optimal initialization parameters and the most suitable network structure configuration. Then, for the calculation of the classification result, a multi-region test method is used, and the accuracy of image recognition can be improved by calculating a plurality of regions of the image during the test. Finally, a general dataset input interface is designed for the system, and the experimental results show that the improved convolutional neural network structure is more conducive to obtaining the diversity characteristics of the image. Under the same experimental conditions, the recognition rate is higher than that of the traditional network.