Aiming at the application of the overhead transmission line insulator patrol inspection requirements based on the unmanned aerial vehicle (UAV), a lightweight ECA-YOLOX-Tiny model is proposed by embedding the efficient channel attention (ECA) module into the lightweight YOLOX-Tiny model. Some measures of data augmentation, input image resolution improvement and adaptive cosine annealing learning rate are used to improve the target detection accuracy. The data of the standard China power line insulator dataset (CPLID) are used to train and verify the model. Through a longitudinal comparison before and after the model improved, and a cross-sectional comparison with other similar models, the advantages of the proposed model are verified in terms of multi-target identification for normal insulators, localization for small target defect areas, and the parameters required for calculation. Finally, the comparative analysis between the proposed ECA-YOLOX-Tiny model and YOLOV4-Tiny model is given by introducing the visualization method of class activation mapping (CAM). The comparative results show that the ECA-YOLOX-Tiny model is more accurate in locating the self-explosion areas of defective insulators, and has a higher response rate for decision areas and some special backgrounds, such as the overlapping small target insulators, the insulators obscured by tower poles, or the insulators with high-similarity backgrounds.

In order to meet the application requirements on unmanned aerial vehicle (UAV) for daily inspection of insulators in overhead grid transmission lines, a lightweight GhostNet-YOLOV4 model is proposed to identify the insulators objects and detect their self-explosion defect meanwhile. In the proposed model, the lightweight GhostNet with full convolutional attention module (C-SE) model is embedded into the YOLOV4 as backbone feature extraction network, and the ordinary convolution and Mish activation function of enhanced feature extraction network in GhostNet-YOLOV4 are replaced by the depth separable convolution and ReLU6 activation function respectively. Meanwhile, the combination methods of data augmentation, label smoothing, K-Means algorithm, adaptive cosine annealing learning rate, and Focal Loss function are adopted to optimize the model. Finally, the class activation mapping (CAM) is used to visually analyze for the identification process of the proposed model. The example analysis results show that the object recognition accuracy mAP of the proposed model is 99.50% for insulator objects and self-explosion defect, and the image processing speed FPS is 2.53 frames/s. The proposed model meets the requirements of lightweight, detect accuracy and speed for UAV application.