In the process of turning processing, tool wear is one of the most important factors affecting machining efficiency, before excessive damage to the work piece surface, it is necessary to identify and update the wear of the tool in time to achieve high-quality production and processing of the work piece. In this paper, a tool wear state recognition method based on deep learning is proposed, which records the turning tool images at different wear stages through a microscope, and uses the convolutional neural network to extract the different wear state features of the input images, and selects appropriate training parameters for the model to realize the state classification of cutting tool wear. Experiments show that the accuracy rate can reach 94.0% when classifying different wear states of turning tools, which can be used to identify tool wear states in the turning process at low cost.