Abstract: Image segmentation of mine personnel is one of the important tasks to realize the technology of personnel detection,behavior recognition,video locating and tracking.However,due to the special underground environment in coal mine,it is difficult for conventional methods to meet the requirement of accurate segmentation of underground personnel.To solve the segmentation problem of personnel image in coal mine,a segmentation method based on superpixel granulation and the clustering of homogenous image granules is proposed,which is appropriate for the personnel images in various scenarios.Firstly,the simple linear iterative clustering (SLIC) model is employed to initially segment the personnel image in coal mine into superpixels units,moreover,the personnel superpixels are identified by the RGB similarity relationship between the superpixels and the marked personnel pixels in sample images.Secondly,the under-segmented personnel superpixels are detected and thoroughly segmented into two with the guidance of their neighbor superpixels,one is elite personnel superpixels and whose texture and grayscale features are extracted.Thirdly,the adjacent elite personnel superpixels with the most similar image features are defined as homogenous image granules,which merge with each other and cluster to generate a homogenous personnel region with specific semantic information.Finally,all the homogenous personnel regions together constitute the whole personnel region,and the personnel region is separated from the background.Personnel images of underground coal mine with four different scenarios are used to verify the performance of the proposed algorithm.The experimental results show that the F-measure value of the proposed algorithm of superpixel granulation is 2.11%,3.36%,13.16%,and 6.82% higher than the average value of the comparison algorithm,and the accuracy value of the proposed algorithm of clustering of homogeneous personnel image granules reaches up to 99.0%,100%,94.4% and 93.75% respectively.In addition,the proposed segmentation method has strong robustness and good segmentation effect for all personnel images in four different mine scenarios.