基于径向基概率神经网络的输电导线缺陷状态识别

输电导线作为承担电能传输任务的重要部件,及时发现其本体缺陷对指导维修避免重大电力事故的发生具有重要意义。考虑到无人机巡检中输电导线背景的复杂性和导线表面缺陷检测的困难度,提出一种基于径向基概率神经网络的输电导线缺陷状态识别方法。首先,依次采用加权色差法、最大类间方差法以及形态学滤波实现复杂背景下输电导线的准确分割。其次,将分割出的导线区域等距划分为10个导线子图像,通过Gabor滤波器获得输电导线8个角度、5个尺度的40幅纹理增强子图像,提取各个子图像的粗糙度、对比度和方向度3个纹理特征量,结合特征方差比筛选出10个强纹理特征;最后,将10个强纹理特征量作为径向基概率神经网络的输入,完成输电导线缺陷状态的识别。实验结果表明所提方法可以实现复杂背景下输电导线快速分割与缺陷状态的准确识别,为无人机巡检中输电导线的运行状态检测提供了新的思路。

State Identification of Transmission Line Defect Based on Radial Basis Probabilistic Neural Network

The transmission line plays an important part in power transmission task, so it is of great significance to identify its defects for the maintenance, and the severe power accidents can be avoided or decreased. For the background of images captured by unmanned aerial vehicle is very complex and difficult to be detected, a radial basis probabilistic neural network based fault location identification method for transmission lines is proposed. Firstly, the weighted color difference method, maximum interclass variance method and morphological filtering are sequentially adopted to realize the accurate segmentation of transmission lines in complicated background. Secondly, the segmented line area is equally divided into 10 line sub-images, 40 texture enhancement sub-images at 8 angles and 5 dimensions of transmission lines are obtained by Gabor filter, and the roughness, contrast and orientation of each sub-image are also extracted. By the feature variance, 10 strong texture features are selected and adopted as the input parameters to the radial basis probabilistic neural network for the defect identification of transmission line. The results show that both the rapid segmentation of transmission lines and the accurate identification of the defects based on the images in the complex background can be achieved by the proposed method, which provides a new idea for the operation state detection of transmission line in unmanned aerial vehicle inspection.