Abstract: The face image information of coal preparation plant is easily affected by complex environmental factors, which makes the recognition difficult. In order to solve this problem, a face recognition method in working environment of coal preparation plant is proposed. The Gabor wavelet transform is applied to the normalized original face image of coal preparation plant to obtain characteristic maps of 8 directions and 5 scales. The method encodes with the improved AR-LGC coding algorithm, and characteristic fusion is performed on the encoded maps of different directions at the same scale to obtain the fused characteristic map of the image. The fused characteristic map is divided into multiple sub-blocks, and the histogram characteristic vector H is obtained by counting the block histogram and weighting cascade. Then H is trained in residual neural network to realize the face recognition of the personnel in the coal preparation plant. The improved AR-LGC coding algorithm enhances the texture correlation of face images in coal preparation plant, solves the problem of insufficient image texture correlation, retains more important characteristics in face images while weakening the interference characteristic, and alleviates the problem of personnel faces being polluted by coal ash. The experimental results show that when the faces of coal preparation plant are polluted by coal ash, the characteristics extracted by the improved AR-LGC coding algorithm retain the local characteristics coarse granularity and have better noise immunity. The recognition rate of the improved AR-LGC coding algorithm is 94.5% and the average time consumed is 0.933 0 s. Compared with similar algorithms, the recognition rate of this algorithm has improved under the condition of sacrificing part of the time performance, and the sacrificed time performance is acceptable.