Research on Recognition Method of Electrical Components Based on FEYOLOv4-tiny

Recently, electrical component recognition technology is of great significance for fault identification and stable operation of the modern power grids. With the rapid improvement of smart grids, higher requirements are put forward to the recognition methods in detection accuracy and real-time performance. However, the conventional recognition methods are fail to meet the demands since they always have drawbacks in detection performance. To improve the electrical component identification capability, a novel detection method based on Feature Enhanced You Only Look Once v4-tiny (FEYOLOv4-tiny) is proposed in this study. Here, YOLOv4-tiny is employed as the prototype of the proposed network, whereas enhancement module is designed to improve the feature extraction capability. Moreover, frequency channel attention and spatial attention module are utilized to capture the informative and discriminatory features. Experimental results indicate that our proposed method outperforms other lightweight networks in detection accuracy with almost the same real-time performance, especially for the small targets with complex background. Over all, FEYOLO v4-tiny is efficient in electrical component identification and has significant application prospects in power inspection.