Wire recognition method based on image recognition

Abstract

At this stage, the detection method of UAV carrying tools has become an indispensable means of maintenance for wire identification. The results of traditional detection methods are not intuitive or the false detection rate is high. For the above problems, this paper proposes a wire identification method based on lightweight Yolov4. Firstly, MobileNetv2 is used as the lightweight backbone feature network, and Sandglass Block is used to reduce the loss of feature information. Then, the Convolutional Block Attention Module (CBAM) is added to improve the accuracy of small target recognition. Finally, the target of the overhead transmission line is identified by judging whether the insulator and the overhead transmission line exist together in the image. The experimental results show that the mAP of the improved method is 96.78%, the FPS is 87.74, and the model size is only 22.74MB. The proposed method can satisfy the small equipment's identification of overhead transmission lines, and the error detection rate is low.