Abstract: In view of problems of low recognition rate and complex algorithm of traditional miner's facial expression recognition methods, based on convolutional neural network and combining with nonlinear mapping function in support vector machine algorithm, a miners' facial expression recognition method based on convolutional neural network was proposed. The convolutional neural network adopts sharing weights strategy, constructs convolutional layer directly with fixed weights, and determine network hierarchy according to matching growth rules. Preprocessed miner's facial expression images are used as test set and training sets of the convolutional neural network. Supportive vector machine is used to classify neurons that represent miner's facial expression features, so as to realize classification and recognition of miner's facial expressions. The experimental results show that the recognition rate of miner's facial expression of the proposed method reaches 90.71%, which can meet the practical application needs.