基于迁移学习的坝面表观缺陷智能检测方法研究

针对常规缺陷检测方法难适用于复杂环境下的坝面表观缺陷检测的问题,提出了一种基于迁移学习的坝面表观缺陷智能检测方法,主要解决坝面缺陷的识别与分类问题。该检测方法主要包括三个部分:首先采用图像预处理对多旋翼无人机采集到的原始图像数据进行数据扩充和特征突显;然后运用迁移学习方法将Inception-v3网络模型作为预训练模型,训练处理过后的缺陷数据,得到坝面缺陷检测模型;最后构建全连接分类网络并利用检测模型对测试集数据进行分类测试。试验结果显示:该检测方法仅耗时28 min就完成了对约33 000张缺陷数据的训练与测试,并对混凝土坝面存在的裂缝、漏筋、渗水和脱落四种缺陷的分类正确率达到了96%。结果表明,该检测方法能够实现对坝面缺陷精确且快速的识别和分类,能够为坝面后期的风险评估和维护提供有力的数据支撑,具有一定的工程意义。

Transfer learning-based study on method of intelligent detection of dam surface apparent defect

Aiming at the problem that the conventional defect detection method is difficult to be applied to the detection of the apparent defect on the dam surface under complicated environment，a transfer learning-based intelligent detection method for the apparent defect of dam surface is proposed herein for mainly solving the problem of identification and classification of dam surface defect. The detection method mainly includes three parts: data expansion and feature highlighting of the original image data co-llected by the multi-rotor UAV are made through image preprocessing at first; and then the Inception-v3 network model is taken as a pre-training model by means of migration learning method for training the processed defect data to get dam surface defect detection model; finally，the fully connected classification network is constructed and the testing set data are classified and tested with the detection model. The experimental results show that the training and testing of about 33 000 defect data are finished within only 28 min with the classification accurate rates of 96% for the four defects on concrete dam surface，i. e. cracks，exposed reinforcing bar，seepage and delamination. The study result shows that this detection method can not only realize the accurate and quick identification of dam surface defect，but also can provide powerful data support for the later-stage risk assessment and maintenance of dam surface，thus has a certain engineering significance.