混凝土裂缝无损检测的改进ResNet方法

Non-destructive detection of concrete cracks based on improved ResNet method

大型混凝土结构的灾难性事故多由微小裂缝发展而成,在混凝土结构服役期间对其进行裂缝检测十分重要。目前基于深度学习的混凝土裂缝无损检测算法飞速发展,但大多未考虑裂缝信息的本身特点,检测的准确性仍有进一步提升空间。为此,提出了一种针对混凝土裂缝无损检测的改进ResNet方法,以残差神经网络ResNet为裂缝检测的基础模型,插入注意力机制模块,提高模型表征能力,使其能够有效捕捉裂缝图像中的重要特征信息,从而提高检测的准确性和鲁棒性。同时,采用迁移学习的策略,将ResNet模型在复杂数据集上的训练成果迁移到裂缝数据集,节约了训练时间和计算资源。结果表明:改进后的ResNet算法的裂缝检测准确率高达98.80%,比原始ResNet算法准确率提升了3.24%。相关经验可供类似改进算法的构建参考。

The catastrophic accidents of large concrete structures are mostly caused by small cracks,so it is very important to detect cracks during the service life of concrete structures.At present,the non-destructive detection algorithm for concrete cracks based on deep learning is developing rapidly,but most of them do not take the characteristics of crack information into account,so the detection accuracy still has room for further improvement.In this paper,an improved ResNet method for non-destructive detection of concrete cracks was proposed.Namely the residual neural network ResNet was used as the basic model of crack detection,and the attention mechanism module was inserted to improve the model representation ability,so that it can effectively capture the important feature information in the crack image,and improve the accuracy and robustness of the detection.At the same time,the transfer learning strategy was used to transfer the training results of the ResNet model on the complex data set to the crack data set,saving training time and computing resources.The results showed that the accuracy of the improved ResNet algorithm for crack detection was as high as 98.80%,which was 3.24%higher than that of the original ResNet algorithm.The relevant experience can be a reference in the construction of similar improved algorithms.