基于机器视觉的填埋场防渗层破损识别方法

填埋场防渗层高密度聚乙烯（HDPE）膜在运营中极易破损，运用在线监测技术确定渗漏区域，将该区域膜上介质移除后，需对漏洞边缘进行精确识别，为实现智能焊接提供视觉基础。因此，提出一种基于机器视觉的填埋场防渗层破损识别方法。首先对样本集进行图像处理，包括图像灰度化、高斯滤波除噪、点运算增强、阈值法分割以及数学形态学处理；其次根据图像的形态特征提取连通域数量、破损面积、周长、长轴、短轴以及轴比，采用holdout方法将样本集划分为训练集与测试集，并将提取到的特征作为输入量，对SVM进行训练；最后采用多个SVM进行分类识别。经实验验证，分类器的总体识别准确率为98.33%，其中块状破损识别准确率为98.24%，缝式破损为98.42%。

RESEARCH ON THE DAMAGE RECOGNIZING METHOD OF IMPERVIOUS LAYER OF LANDFILL BASED ON MACHINE VISION

The high-density polyethylene (HDPE) film of the anti-seepage layer of the landfill is easily damaged during operation. The online monitoring technology is used to determine the leakage area. After the medium on the membrane removed, the loopholes need to be accurately identified to provide a visual basis for welding process. Therefore, a machine vision-based damage identification method for impermeable layer in landfill was proposed. First, perform image processing on the sample set, including image grayscale, Gaussian filter denoising, point operation enhancement, threshold segmentation, and mathematical morphology processing. Secondly, the number of connected domains, damage area, circumference, major axis, minor axis and axial ratio were extracted according to the morphological features of the image. The retention method weas used to divide the sample set into a training set and a test set, and then the extracted features were used as the input for training SVM. Finally, multiple SVMs were used for classification and recognition. Experiments showed that the overall recognition accuracy of the classifier was 98.33%, among which the accuracy of block damage recognition was 98.24%, and the stitch damage was 98.42%.