There are many kinds of fish in China, and their shape is an important basis for their classification. In order to solve the difficulty in artificial fish identification, an intelligent fish identification system based on deep learning has been proposed to realize the intelligent identification of 1400 species of fish in China. The system first uses convolutional neural network’s EfficientNet model to train a dataset of 500,000 pictures of 1,400 species of fish. The accuracy of the model is 95%, and the recognition time of a single picture is only 0.2 seconds, the model’s size is 74.5 MB. Then the front-end of the system uses Wechat applet, the back-end uses Spring + Spring MVC + Mybatis SSM architecture, invoking the training of the model file for identification, the realization of fish identification, page rendering, statistical analysis and recommendation of adjacent species. The system’s design and implementation  provide a feasible idea for the application of intelligent fish identification technology to mobile terminal, which   marine researchers and developers can use as reference.