CHARACTER MODEL TRAINING METHOD, CHARACTER RECOGNITION METHOD, APPARATUSES, DEVICE AND MEDIUM

摘要

A character model training method, a character recognition method, apparatuses, a device and a medium. The character model training method comprises: acquiring a handwritten character training sample, the handwritten character training sample comprising handwritten character images and labeled Chinese characters (S10); if the number of the handwritten character images is greater than a preset number threshold, randomly selecting, from the handwritten character training sample, the same number of handwritten character images as the number threshold; inputting the selected handwritten character images into a convolutional cyclic neural network model; using a random gradient descent-based reverse propagation algorithm to update a weight value and an offset in the convolutional cyclic neural network model, so as to acquire a handwritten character training model (S20); if the number of the handwritten character images is not greater than the preset number threshold, inputting all the handwritten character images into the convolutional cyclic neural network model; using a batch gradient-based reverse propagation algorithm to update the weight value and the offset in the convolutional cyclic neural network model, so as to acquire a handwritten character training model (S30); acquiring a handwritten character test sample, inputting the handwritten character test sample into the handwritten character training model to acquire a recognition accuracy rate, if the recognition accuracy rate is equal to or greater than a preset accuracy rate, determining that the handwritten character training model is a handwritten character recognition model (S40). The present invention ensures the accuracy of training of the handwritten character training model, without affecting the training speed of the handwritten character training model, so that the handwritten character recognition model accurately recognizes a handwritten character.