Defect Identification of Power Line Insulator Based on an Improved yolov4-tiny Algorithm

With the rapid development of artificial intelligence technology, the collection of power line images through unmanned aerial vehicles (UAV) and the further use of deep learning algorithms to automatically detect power insulator defects is gradually replacing the traditional manual inspection and identification methods. The image capture and machine recognition and learning has gradually become a kind of emerging automatic inspection methods of power line insulators. Aiming at the problems of small target size, complex background, and low defect recognition rate of insulator defects, this paper proposes an insulator defect detection method based on an improved yolov4-tiny deep learning algorithm (IYTDLA). Compared to the traditional yolov4-tiny deep learning algorithm (TYTDLA), the key improvement lies that a coordinate attention (CA) module is introduced after the major feature extraction network to enhance the network feature representation ability. First, based on the original UAV-collected image data sets, the improved algorithm is used to randomly scale captured images, and Gaussian noise is mixed to further enhance the data sets. Then, IYTDLA is applied to discern the two different defects “missing” and “broken” from the power line insulator images captured by UAVs. The experimental results show that the IYTDLA has a higher recognition accuracy than the TYTDLA. Compared to the mAP of TYTDLA, the mean average precision (mAP) of IYTDLA is increased by 0.94%, the average precision (AP) of missing insulator defects of IYTDLA is increased by 1.19%, the AP of insulator damage defects of IYTDLA is increased by 2.99%. At the same time, the performances of IYTDLA are also higher than those of traditional faster-rcnn (FRCNN) and efficientdet (EDET) in terms of recognition accuracy and processing speed. However, IYTDLA also has a processing speed comparable to the TYTDLA. That verifies that both of IYTDLA and TYTDLA are suitable for the deploy applications on mobile devices or embedded devices to implement device-side edge computing.