Abstract: Using machine vision technology to extract regions of interest (ROI) from infrared thermal images is a key link for intelligent diagnosis of infrared faults in power distribution equipment. But the mismatch of threshold selection of multi-target region leads to the loss of key details of image. Therefore, this paper puts forward an infrared thermal image multi-threshold segmentation technology of bat bionic algorithm, and uses the optimization threshold to carry on the image segmentation experiment to the infrared thermal image multi-target of power distribution equipment, which breaks through the problem of improving the accuracy index of the algorithm. Compared with the classical image segmentation method, this method can effectively improve the speed and accuracy of image segmentation, and solve the problem of hot image ROI region extraction in complex environment.