Extraction of abnormal heating regions from infrared images is the important prerequisite for intelligence diagnosis of thermal state of electrical equipment. Aiming at cable terminations, an automatic extraction method is proposed in this paper. Firstly, an adaptive wavelet threshold denoising method based on Maximum a Posteriori Estimation (MAP) is applied to remove noise and improve the quality of infrared images. Then, to eliminate the interference, the deep learning network is used to identify and locate the cable terminations in the images. Finally, the Mean-Shift algorithm is applied to cluster the pixels of cable terminations, and extract the abnormal heating regions. Case studies given in this paper show that the proposed method is suitable for infrared images under different backgrounds and different angles. After identifying and locating the cable terminations, the overheating regions can be extracted accurately. Compared with some existing methods, the proposed method has better performance in efficiency and accuracy.