Crack grid detection and calculation based on convolutional neural network

Pavement crack detection has always been an important problem in pavement maintenance. With rapid development in computer vision technology, crack detection using deep learning has become one of the major tasks in automatic pavement condition detection and assessment. This paper aims to develop a method of crack grid detection based on convolutional neural network. First, an image denoising operation is conducted to improve image quality. Next, the processed images are divided into grids of different scales (10 x 10, 20 x 20, 30 x 30), and each grid is fed into a convolutional neural network for detection. The pieces of the grids with cracks are marked and then returned to the original images. Finally, on the basis of the detection results, threshold segmentation is performed only on the marked grids. Information about the crack parameters is obtained via pixel scanning and calculation, which realizes complete crack detection. The experimental results show that 30 x 30 grids perform the best with the accuracy value of 97.33%. The advantage of automatic crack grid detection is that it can avoid fracture phenomenon in crack identification and ensure the integrity of cracks. Thresholding smaller grids allows better combinations of the differences between objects and backgrounds, which makes segmentation results more accurate.

Automated summary

In this paper, crack grid detection based on convolutional neural network is developed, achieving high accuracy crack detection through image denoising and grid segmentation. The 30 x 30 grids perform the best, automatic detection avoids fracture phenomenon, and ensures the integrity of cracks.