Abstract: In order to solve the problems of manual coal preparation and wet preparation method, such as inefficiency, high labor intensity, water consumption, environmental pollution. This paper studies the method of coal gangue recognition based on machine vision, builds a test platform in the laboratory, develops an application platform of MFC software, and realizes the real-time recognition of coal gangue; selects coal and gangue from Shanxi, Inner Mongolia, Shaanxi as samples, and establishes a sample image library; takes 420 images as experimental samples, and extracts the gray mean value, peak gray value, energy, entropy, contrast and deficit of samples The six features of moment are analyzed and counted; the particle swarm optimization (PSO) algorithm is used to optimize the support vector machine (SVM), and the classifier is trained and tested. The results of feature analysis show that gray-scale features have better discrimination than texture features; in PSO-SVM classifier test, when gray-scale, texture and combined features are used as input, the recognition accuracy is 95.83%、72.92%、 and 93.75% respectively, and the results show that gray features are the best input recognition effect.