Infrared thermal image region of interest (ROI) extraction has important significance for fault detection, target tracking and so on. In order to solve the problems of many infrared thermal image disturbances, artificial markers and low accuracy, a ROI of infrared thermal image extraction algorithm based on fusion of multi-modal feature map is proposed. Multi-modal feature maps are constructed by contrast, entropy, and gradient features, and region filling is performed to achieve ROI extraction. New algorithm is applied to actual collected photovoltaic solar panel image. Simulation results show that the proposed algorithm has high average precision (93.0553%), high average recall (902841%), F1 index and J index are better than Grab Cut, less artificial marks, etc.. It can be effectively used for ROI extraction of infrared thermal images.