Abstract: This study aims to address the problems of low accuracy and long processing times of infrared image segmentation of power equipment in complex backgrounds. In this study, an infrared image segmentation algorithm based on the improved DeepLabv3 + algorithm is proposed for power equipment. First, lightweight CA − MobileNetV3 was used instead of Xception to realize feature extraction, reduce model parameters, and improve segmentation accuracy. Second, atrous spatial pyramid pooling (ASPP) was replaced with spatial pyramid (SP)-Dense ASPP to extract a denser and wider range of detailed features and enhance the strip characteristics. Finally, the efficient channel attention (ECA) mechanism was introduced to realize the effective fusion of different levels of feature information and improve segmentation accuracy and robustness of the model. The experimental results showed that the proposed algorithm had higher feasibility and effectiveness in the actual infrared image segmentation task of power equipment than the four more advanced semantic segmentation models. The average increase in mean pixel accuracy (MPA) was 2.67%, and the average increase in mean intersection over union (mIoU) was 9.32%.