基于YOLOv2无人机高压绝缘子实时检测研究

Research on Real-time Detection of UAV High-voltage Insulators Based on YOLOv2

在高压输电线路中高压绝缘子需要不断检查和监测来防止故障和突发事件的发生,然而在绝缘子经常遭受恶劣天气条件的山区,人工检测的成本昂贵。本文利用航空图像中的YOLO(You Only Look Once)深度学习神经网络模型,在背景不杂乱、目标分辨率和光照条件变化的情况下,为绝缘子的检测提供了一种经济有效的解决方案。首先从航拍绝缘子自动检测绝缘子,然后再利用无人机对绝缘子状况进行实时分类,最后运用数据增强方法来避免实验中56000个图像样本的过度拟合。实验证明,该方法能够在无人机实时图像数据上准确定位绝缘子,然后用不同的分类对检测到的绝缘子图像进行冰、雪和水存在下的绝缘子表面状况评估。

In high-voltage transmission lines, high-voltage insulators need to be continuously inspected and monitored to prevent failures and emergencies.However, in mountainous areas where insulators are often subjected to severe weather conditions, manual inspection is expensive.This paper uses the YOLO(You Only Look Once)deep learning neural network model in aerial images to provide an economical and effective solution for the detection of insulators under the condition that the background is not cluttered, the target resolution and the light conditions are changed.First, the insulators are automatically detected from the aerial photography, and then the UAV is used to classify the insulators in real time.Finally, the data enhancement method is used to avoid the over-fitting of the 56000 image samples in the experiment.Experiments have proved that this method can accur-ately locate the insulator on the real-time UAV image data, and then use different classifications to evaluate the insulator surface condition in the presence of ice, snow and water.