基于检测图像的排水管道缺陷智能辅助分类方法

针对当前地下管网CCTV检测缺陷中存在自动化程度偏低及依赖专业人员技术水平的问题，综合采用图像处理和深度学习技术构建了辅助检测人员快速、准确地识别管道缺陷类型的智能方法。首先，收集十类典型缺陷图像，对其进行图像处理生成样本集；在此基础上，以深度卷积神经网络AlexNet和ResNet50为基础框架，使用预训练AlexNet和ResNet50网络迁移学习管道缺陷特征，通过敏感性分析优化了分类网络参数，然后，通过测试集验证了管道缺陷智能分类模型的准确性，并结合具体工程实例验证建立方法的有效性。结果表明：两类管道缺陷智能分类模型在测试集上分别达到92.00%和96.50%的准确率，实际工程实例准确率达到了85.41%和87.94%，且ResNet50的分类效果更优，具有较好工程适应性。图像处理和深度学习技术可提高排水管道缺陷分类的自动化与准确率，值得进一步进行推广。

Research on Intelligent Auxiliary Classification Method for Drainage Pipeline Defects Based on Inspection Images

In view of the problems of low automation and dependence on the technical level of professionals in the current CCTV detection defects of underground pipeline networks, an integrated method of image processing and deep learning technology is used to build an intelligent method to assist the detection personnel to quickly and accurately identify the type of pipeline defects. Firstly, ten kinds of typical defect images are collected and processed to generate a sample set ; on this basis, using deep convolutional neural networks AlexNet and ResNet50 as the basic framework, using pre-trained AlexNet and ResNet50 networks to learn pipeline defect features, through Sensitivity analysis optimizes the classification network parameters. Then, the accuracy of the pipeline defect intelligent classification model is verified through the test set, and the effectiveness of the model is verified with specific engineering examples. The results show that the accuracy rates of two networks achieve 92.00% and 96.50% on the test set, and the accuracy of the actual engineering cases reached 85.41% and 87.94%, and the classification effect of ResNet50 was better, with good engineering adaptability. Image processing and deep learning technology can improve the automation and accuracy of the classification of drainage pipeline defects, and it is worth further promotion.