Abstract: Because of large amount of parameters and high hardware requirements, current deep learning algorithms are difficult to be embedded in drones and other mobile devices. In order to enable the drone to carry a lightweight model and to identify the surface faults of overhead transmission lines insulators, the MobileNet-SSD target detection network and MobileNetV2-DeeplabV3+ image segmentation network are integrated to identify and segment the self-explosion faults off insulators. Based on the network characteristics, the MobileNet-SSD is used to accurately classify and locate the insulators firstly, and then the semantic segmentation algorithms of MobileNetV2-DeeplabV3+ are used to segment the self-explosion pictures of insulator. The example analysis shows that the insulators can be identified quickly and the self-explosion faults of insulators can be segmented accurately based on the proposed method even if the backgrounds are complex. The proposed method has the characteristics of less model parameters, high computing efficiency, strong robustness etc, and it can meet the embedded application requirements. Therefore, the accuracy and real-time of inspection for overhead transmission lines by drones can be improved.