An method for power lines insulator defect detection with attention feedback and double spatial pyramid

In recent years, with the wide application of UAVs in electric power inspection, inspection efficiency is promoted rapidly. However, how to address massive aerial images is a big challenge. Deep learning applied to automated processing and analysis of aerial images has become a hot topic in current research on insulator defect detection. Most of the existing methods based on deep learning are highly sensitive to complex background which makes detection precision of multi-scale insulators and its defects susceptible to a great influence. Moreover, sheltered insulators cannot be detected in current study. In order to improve the detection precision of insulator defects under complex background and detect sheltered insulators, in this paper, we propose a method that applies the Attention Feedback (AF) and the Double Spatial Pyramid (DSP). AF can enhance attention ability of discriminative features. Furthermore, DSP can combine different pooling response values by using a set of scale factors and two pooling ways. The experiment results show that the proposed method for power lines insulator defect detection with Attention Feedback and Double Spatial Pyramid can achieve Precision of 98.9%, Recall of 99.5%, F1-score of 99.0%, and reaches mAP of 97.1%. In complex background images, the method not only can detect insulator defects, but also identify the sheltered insulators.