Abstract. As insulators play an important role in power transmission lines, we propose an intelligent method based on the convolution neural network (CNN) to evaluate the corona discharge of an insulator. In this method, the imaging device is a dual-spectra camera with a visible channel and an ultraviolet (UV) channel. The CNN is adopted to identify the detection distance of the insulator with the visible channel. To train the network, the dataset of the insulator is obtained by the experimental setup and deep convolutional generative adversarial networks. Through adjusting the training parameters and optimizing the network structure, an optimal trained model is achieved. Then the image pixel ratio method is adopted to measure the UV signal strength of the images captured by the UV channel. Meanwhile, the relationship between the detection distance and the UV signal strength is discussed. The critical value for the corona discharge of the insulator is obtained via experiments at the standard detection distance. Finally, the corona discharge of the insulator is evaluated by combining the detection distance with the UV signal strength. The experimental results show the method has the advantages of high accuracy and robustness and can effectively evaluate the corona discharge of the insulator.

中文翻译：

---

基于卷积神经网络的双光谱相机绝缘子电晕放电评价方法

摘要。由于绝缘子在输电线路中起着重要作用，我们提出了一种基于卷积神经网络 (CNN) 的智能方法来评估绝缘子的电晕放电。在该方法中，成像设备是具有可见光通道和紫外 (UV) 通道的双光谱相机。采用CNN识别绝缘体与可见通道的检测距离。为了训练网络，绝缘体的数据集是通过实验设置和深度卷积生成对抗网络获得的。通过调整训练参数和优化网络结构，得到一个最优的训练模型。然后采用图像像素比的方法测量UV通道捕获的图像的UV信号强度。同时，讨论了检测距离与紫外信号强度之间的关系。绝缘子电晕放电的临界值是通过在标准检测距离下的实验获得的。最后，结合检测距离和紫外信号强度来评估绝缘体的电晕放电。实验结果表明，该方法具有精度高、鲁棒性好等优点，能有效评估绝缘子的电晕放电。