Combined with the residual and multi-scale method for Chinese thermal power system record text recognition

Abstract

Aiming at the problem that the recognition accuracy based on convolutional neural network of thermal power system record text is not high, a method of thermal power system record text recognition based on residual and multi-scale feature combination was proposed and implemented. Combined with the residual, a new network is designed to replace the traditional convolutional neural network and improve the feature extraction ability of the network. The 1 ? 1 convolution core was used to increase the network depth and reduce the parameters instead of the 3 ? 3 convolution core. The network order of each layer in residual block was adjusted so that the network representation ability can be improved. Combining feature information of different scales and retaining more vertical feature information, the classification accuracy of the network is improved. Experiments on the self-built image data set of thermal power system records show that the proposed network model improves the accuracy by 11% compared with convolutional recurrent neural network, and has better robustness to image distortion and blurring.