Abstract: For the photovoltaic infrared hot spot detection problem,this paper proposes a curve fitting combined with image clustering of the hot spot infrared image processing methods. Firstly,after image gray scale transformation,Gaussian least square fitting was used to determine the clustering center. In view of the poor robustness of traditional FCM noise,the hot spot images were clustered by adding the influence of neighborhood space and substituting the Euclidean distance with the kernel distance. Finally,the gray multi-threshold segmentation was carried out according to the fitting graph. The experimental results show that the method can quantify the damage degree of photovoltaic modules,regional stratification,suppress infrared image noise,improve the efficiency of hot spot detection with the segmentation accuracy of more than 86%.