CHARACTER RECOGNITION NETWORK MODEL TRAINING METHOD, CHARACTER RECOGNITION METHOD, APPARATUSES, TERMINAL, AND COMPUTER STORAGE MEDIUM THEREFOR

摘要

A character recognition network model training method, a character recognition method, apparatuses, a terminal, and a computer storage medium therefor. The character recognition method comprises: standardizing a picture to be tested, and scaling same to a preset height H and a preset width W (A100); inputting said picture into a convolutional neural network, and extracting a convolutional feature of said picture, so as to obtain a depth feature map that includes the convolutional feature (A200); inputting the depth feature map into an attention mechanism module provided with multiple channels to obtain an attention weight of each channel, and rescaling each channel of the depth feature map by using the attention weight to obtain multiple attention feature maps (A300); respectively inputting each of the attention feature maps into a fully connected layer to obtain multiple attention feature vectors (A400); and performing feature fusion on the multiple attention feature vectors, and inputting same into a character category fully connected layer to perform character category prediction (A500).