基于改进ART-2网络和不变矩的高压瓷瓶裂缝识别

Recognition of porcelain bottle crack based on modified ART-2 network and invariant moment

为了保证高压输电线路的正常运行,可以通过高压输电线路巡检机器人视觉系统完成高压输电线路的检测。本文通过CCD摄像头等硬件模拟机器人的视觉,完成对绝缘瓷瓶裂缝图像的采集。对图像经过滤波去噪、图像分割等预处理操作后,利用形状特征和灰度差异完成图像中裂缝的定位。对于聚焦放大后的裂缝图像提取不变矩等四个特征值,得出图像信息。最后利用改进的ART-2神经网路,实现对绝缘瓷瓶裂缝五种状态:横向、纵向、块状、网状、无裂缝的分类识别。通过仿真和实验表明该算法可以有效、可靠地运用于绝缘瓷瓶裂缝类型识别研究中,并可方便地应用于其它领域。

In order to ensure the security of power transmission lines, the power transmission lines can be inspected using vision system of an inspection robot. In this paper, the images of porcelain bottle crack are collected by some hardware such as CCD, which is used to simulate the vision system of the robot. After image preprocessing, such as smoothing and segmenting the objects from the background with a threshold, the eigenvector and gray difference of the image are used to obtain the orientation of the crack in the image. Four features, such as invariant moment and so on, are extracted from the image, which can reflect the information of the whole image. Finally, an algorithm is designed using modified ART-2 neural network, which is used to realize the classification of the cracks. The porcelain bottle cracks are classified into five classifications： transverse crack, longitudinal crack, block crack, alligator crack and non-distress crack. The results of simulation and experiment show that the algorithm can be used effectively and reliably in the recognition of porcelain bottle crack types and other fields.