Pixel-level intelligent recognition of concrete cracks based on DRACNN

摘要

Materials surface damage identification based on computer vision technology has become one of the research hotspots in the field of materials surface. Crack is one of the most common forms of material damage. It is of great significance to carry out intelligent recognition of cracks to identify and estimate evolution of material damage. In order to improve the accuracy of intelligent crack recognition, a deep residual attention convolution neural network (DRACNN) was proposed for semantic segmentation of concrete cracks. DRACNN network is based on U-Net and adds recursive residual convolution block and attention mechanism in U-Net for more accurate intelligent crack recognition at pixel-level. Through the comparison with other mainstream semantic segmentation algorithms, it is found that the proposed DRACNN can achieve better classification performance for concrete cracks, and the IoU, accuracy, precision, and recall of the DRACNN are 73.95, 97.82, 78.48, and 67.95 respectively.