Method for Mechanized Construction Erosion Area Identification in Transmission Line Based on Deep Learning

• Published in: 2023 5th International Conference on Electrical Engineering and Control Technologies (CEECT) pp. 634 - 638
• Main Authors: Feng, Jia, Chen, Bin, Li, Enyang, Zhang, Lijun, Liu, Fei, Guo, Chenxin
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.12.2023
• Subjects: Biological system modeling; convolutional block attention module; Erosion; mechanized construction; Neck; Power system stability; Power transmission lines; Soil; soil erosion; transmission line; YOLO; YOLOv5s
• DOI: 10.1109/CEECT59667.2023.10420657

As mechanized construction activities escalate in transmission line projects, the stability of the power system and environmental protection face challenges from soil erosion. This study proposes an improved method for identifying erosion areas using YOLOv5s. It reduces non-essential feature learning costs by introducing the GhostConv module in the Backbone network and enhances the ability to extract exposed soil features by incorporating the convolutional attention module in the Neck network. Finally, it introduces the SIOU loss function to improve the fitting of detection boxes. The improved identification method has a smaller model size, suitable for transmission line scenarios, and achieves an average precision of 95.8% for erosion area detection, 1.5% higher than the original model. The proposed enhancements have been validated through experimental results, affirming their effectiveness and superiority. This study is expected to provide essential support for the early identification of erosion hazards and help maintain the natural environment around transmission lines.