Abstract: In the process of hot spot detection with infrared thermal images, the thermal images will be interfered by many factors introduced by complex environment and acquisition process, resulting in the inability of conventional image processing methods to accurately detect hot spots. An improved hot spot detection method is proposed in this paper, which fits the histogram using the B-spline of adaptive knots according to the grayscale histogram features, so as to realize threshold selection and hot spot segmentation. The method not only has the same characteristics of suppressing noise, but also improves the accuracy of hot spot detection. Experimental results show that this method can realize the hot spot detection of infrared thermal images of photovoltaic modules containing a lot of noise and improve the efficiency and accuracy of hot spot detection.