Comparison of deep convolutional neural networks and edge detectors for image-based crack detection in concrete

Abstract This paper compares the performance of common edge detectors and deep convolutional neural networks (DCNN) for image-based crack detection in concrete structures. A dataset of 19 high definition images (3420 sub-images, 319 with cracks and 3101 without) of concrete is analyzed using six common edge detection schemes (Roberts, Prewitt, Sobel, Laplacian of Gaussian, Butterworth, and Gaussian) and using the AlexNet DCNN architecture in fully trained, transfer learning, and classifier modes. The relative performance of each crack detection method is compared here for the first time on a single dataset. Edge detection methods accurately detected 53–79% of cracked pixels, but they produced residual noise in the final binary images. The best of these methods was useful in detecting cracks wider than 0.1 mm. DCNNs were used to label images, and accurately labeled them with 99% accuracy. In transfer learning mode, the network accurately detected about 86% of cracked images. DCNNs also detected much finer cracks than edge detection methods. In fully trained and classifier modes, the network detected cracks wider than 0.08 mm; in transfer learning mode, the network was able to detect cracks wider than 0.04 mm. Computational times for DCNN are shorter than the most efficient edge detection algorithms, not considering the training process. These results show significant promise for future adoption of DCNN methods for image-based damage detection in concrete. To reduce the residual noise, a hybrid method was proposed by combining the DCNN and edge detectors which reduced the noise by a factor of 24.