为节省人工识别裂缝的时间和提高混凝土检测效率, 通过MATLAB建立卷积神经网络图像检测模型, 探讨研究了图片重构采样率、二值化阈值和3种检测算子(拉普拉斯算子、索贝尔算子和Canny算子)对混凝土裂缝检测的影响。通过调整不同的参数得出: 对于混凝土裂缝图片, 重构图片的采样率为0.3时达到最优; 二值化阈值为0.4时达到最优。Canny算子对边缘检测和裂缝检测的效果最佳, 能够全面地反应出边缘裂缝情况, 索贝尔算子的效果次之, 拉普拉斯算子的效果较差。所建立的卷积神经网络图像检测模型能够为今后复杂环境下混凝土结构检测提供技术方法。

In order to save the time of artificial crack identification and improve the efficiency of concrete detection, the convolution neural network image detection model is established by MATLAB. The influence of image reconstruction sampling rate, binary threshold and three detection operators (Laplace operator, Sobel operator and Canny operator) on concrete crack detection are studied. By adjusting different parameters, it can be concluded that the optimal sampling rate is 0.3 and the optimal threshold is 0.4 for the concrete crack image. Canny operator has the best effect on edge detection and crack detection, and can comprehensively reflect the edge crack situation, followed by Sobel operator, the effect of Laplace operator is poor. The convolution neural network image detection model can provide technical methods for the detection of concrete structures in complex environment in the future.