Abstract: The classification algorithm of Turing pattern with the help of Gray-Scott reaction-diffusion model was investigated. The linear stability theory was used to analyze the coexistence balance point of the system and ensure the different types of patterns produced by Turing instability. Convolutional neural networks instead of multi-layer algorithm feature engineering were adopted to classify the generated patterns. Data features X - X^3 and \nabla X that are more meaningful than X was used to represent the interface and main area of complex pattern data. Therefore, X - X^3 and \nabla X were employed to realize the classification in neural networks and clustering algorithms. It was found that the effectiveness of the method was verified by the results.