This study introduces a novel methodology for electrical discharges recognition elaborating an algorithm based on the RGB colour image model and artificial neural network (ANN) classifier. The developed RGB-ANN algorithm aims to detect and monitor the propagation of electrical discharges until flashover, through analysis of six colours appearing in the discharges images, extracted from the flashover videos recorded on a plan glass insulator model under uniform pollution. First, more than 300 colours images are collected and divided into sets to form a large database. Using an RGB encoding system, each pixel is represented by (R, G, B) coordinates and each image is encoded by 3D matrix. For the discharge image, the coordinates of each pixel are compared to all database ones. The colour of the database having the same coordinates of the discharge image pixel is attributed to this latter. Based on the ratio of the pixels number of a given colour to the total pixels number of the discharge image, six indicators are quantified and grouped to form a feature vector. This latter is used as input of the ANN, in order to classify the evolution of discharges into five classes. As the main result, >98% of images have been well classified.

中文翻译：

---

结合人工神经网络的彩色图像编码系统在绝缘子模型上的放电模式识别

这项研究介绍了一种新颖的放电识别方法，该方法阐述了一种基于RGB彩色图像模型和人工神经网络（ANN）分类器的算法。所开发的RGB-ANN算法旨在通过分析放电图像中出现的六种颜色来检测和监视放电放电，直到放电为止，该六种颜色是从均匀污染的平面玻璃绝缘子模型上录制的闪络视频中提取的。首先，收集了300多种彩色图像并将其分成几组以形成一个大型数据库。使用RGB编码系统，每个像素由（R，G，B）坐标表示，每个图像由3D矩阵编码。对于放电图像，将每个像素的坐标与所有数据库的坐标进行比较。具有与放电图像像素相同的坐标的数据库的颜色归因于后者。基于给定颜色的像素数与放电图像的总像素数之比，对六个指标进行量化和分组以形成特征向量。后者用作人工神经网络的输入，以便将排放的演变分为五类。主要结果是，> 98％的图像已被很好地分类。