【摘要】 混凝土缺陷普遍存在于混凝土建筑物及构筑物中,这些缺陷影响结构美观和使用功能,甚至可能带来结构安全问题。为了解决人工检测存在的费时、费力、危险、易出错等问题,同时提高缺陷检测效率及准确率,降低检测成本,本文针对混凝土表面裂缝和孔洞这两种常见缺陷,提出一种基于图像的混凝土表面缺陷检测方法。该方法结合数字图像处理和深度学习,能同时检测混凝土表面裂缝和孔洞。首先在实验室条件下制备混凝土块采集裂缝和孔洞图像作为样本数据集;随后进行图像预处理,包括RGB三通道阈值分割、形态学处理,以减少原始图片的噪声,提高裂缝及孔洞与背景对比度;最后建立深度卷积神经网络,经测试,裂缝检测正确率为97.48%,孔洞检测正确率为94.08%,裂缝和孔洞测量结果与实际结果误差小于6%。结果表明,本文提出的方法能有效地检测混凝土表面裂缝和孔洞,检测正确率较高,具有良好的应用前景。更多还原

【Abstract】 Concrete defects exist in concrete buildings and structures commonly. These defects affect the structural aesthetics and using functions,and may even bring structural safety problems. In order to solve the problems of being time-consuming,laborious,dangerous and error-prone of manual detection,and to improve the efficiency and accuracy of defect detection,and to reduce the detection cost,this paper proposed an image-based method for the detection of concrete surface defects such as cracks and bugholes. Combined with digital image processing and deep learning,this method could simultaneously identify cracks and bugholes on concrete surface. Firstly,some concrete blocks were prepared under laboratory conditions and images of cracks and bugholes were collected as sample data sets.Then the image preprocessing,including RGB three channel threshold segmentation and morphological processing,was carried out to reduce the noise of the original image and improve the contrast of cracks and holes with the background. Finally,the deep convolutional neural network was established.Through testing,the accuracy of crack identification was 97.48%,and that of bughole identification was 94.08%. The error between the measured results of cracks and bugholes and the actual results was less than 6%. The results show that the method proposed in this paper can effectively detect the cracks and bugholes on the concrete surface,and the detection accuracy is high,which has a good application prospect.更多还原