基于DenseNet分类的隧道裂缝检测研究

Study on Tunnel Crack Detection Based on DenseNet Classification

摘要:

针对隧道裂缝人工识别低效、检修不便以及隧道环境复杂多变、检测易受噪声干扰等问题，文中提出一种基于深度学习的裂缝检测算法。通过神经网络对原始图像进行非裂缝区域过滤，减少无关背景信息的干扰，同时在分割算法基础上通过多维分类器将误识别的裂缝区域剔除。实验结果表明，密集连接卷积网络（DenseNet）在裂缝分类中最高可达99.95%的准确率，有效提升了隧道裂缝自动检测精度。

Abstract:

Aiming at the problems of inefficient manual detection of tunnel cracks, inconvenient maintenance, complicated and changeable tunnel environment, and susceptibility to noise interference, a crack detection algorithm based on deep learning is proposed. Non-crack areas are filtered through the neural network to reduce the interference of irrelevant background information. At the same time, based on the segmentation algorithm, mis-recognized crack areas are eliminated. Experimental results show that the DenseNet network can reach a maximum accuracy of 99.95% in crack classification, which effectively improves the accuracy of tunnel crack detection.