基于YOLO目标检测算法的输电线路检测

Transmission Line Detection Based on YOLO Object Detection Algorithm

输电线路是电网的重要组成部分,确保输电线路安全是保障电力系统稳定运行的重要措施。文章提出一种基于YOLO(you only Look once)算法的输电线路检测模型,用以对无人机等可见光拍摄图像中的输电线路进行物体检测。输电线路的识别检测任务一直是目标检测任务的难题,针对输电线路细长物理结构的特点,对微型目标检测YOLO模型进行改进,设计了对细长物理结构的高长宽比预测锚(anchor),对特征提取网络Darknet进行精简,在保持模型性能和检测精度的基础上对其进行轻量化处理,设计一种可通过单GPU(graphics processing unit)训练的轻量级模型并通过实验与原始YOLO模型、Faster R-CNN以及SSD在测试集上进行性能对比。实验结果表明,轻量级YOLO模型在测试集上检测精度达到83.04%,说明预测锚结构和精简后的网络模型对于输电线路检测是有效的。

Transmission line is an important part of power grid. To ensure the safety of transmission line is an important means to ensure the stable operation of power system. In this paper, a transmission line detection model based on YOLO(you only look once) algorithm is proposed to detect transmission line objects in visible light images such as UAV. Transmission line identification and detection task has always been a difficult problem in the object detection task. In view of the characteristics of the transmission line slender physical structure, this paper improves the YOLO model of the micro object detection, designs an anchor for the slender physical structure, simplifies the feature extraction network Darknet, and on the basis of maintaining the performance and detection accuracy, lightweight processing of the model is carried out to obtain a lightweight YOLO model that can be trained by a single GPU. Through experiments, the performance of lightweight YOLO model is compared with the original YOLO model, Faster R-CNN and SSD on the test set, the results show detection accuracy of the lightweight YOLO model reaches 83.04%, and the prediction anchor structure and the simplified network model proposed in this paper are effective for transmission line detection.