Abstract: In order to improve the stability and accuracy of coal and gangue separation, a multi-feature fusion method based on Fuzzy Support Vector Machine with Normal Plane membership function for Particle Swarm Optimization algorithm (PSO-NP-FSVM) is proposed. The basic principle and workflow of coal and gangue recognition by X-ray detection technology are introduced.After the collected X-ray image pre-processing by median filtering, the gray mean, gray variance under grayscale characteristic and the energy, correlation, contrast, and entropy under the texture characteristic based on the gray level co-occurrence matrix of a total of six characteristic quantities of coal and gangue are extracted respectively. And the selected characteristics is fused. The merits of isolated samples can be effectively eliminated by using the Normal Plane membership function, and the main parameters of the Fuzzy Support Vector Machine classifier model are optimized by combining the Particle Swarm Optimization algorithm. An improved PSO-NP-FSVM classification algorithm is proposed.Using the same training samples, the simulation results are compared with PSO-FSVM classifier model. Finally, the PSO-NP-FSVM, PSO-FSVM and single grayscale characteristic or texture characteristic are used to establish the classifier model, and the cross-validation method is used to carry out comparative experiments. The experimental results show that the PSO-NP-FSVM algorithm achieves the optimal parameters after 56 iterations, and the PSO-FSVM algorithm needs to be iterated 63 times. The PSO-NP-FSVM algorithm has a small fitness function value. And the classification accuracy of coal and gangue is 93.8% through the multi-feature PSO-NP-FSVM method. The accuracy and stability of the new classifier model and single characteristic recognition are better than the common PSO-FSVM classifier model. The photoelectric intelligent separation technology of X-ray detection is an important trend in the future development of coal and gangue sorting. This method can improve the defect of recognition accuracy caused by the influence of thickness for coal and gangue during the sorting process.