Crack detection and recognition based on improved BiSeNetV2

Cracks, as a relatively common discontinuity in solid materials, are the beginning of damage to solid structures. Their timely identification and detection, followed by the analysis of the test results and corresponding measures, can better prevent accidents, and ensure safety in engineering operations. At present, crack identification mainly relies on manual detection, which has the problems of high labor intensity, time consuming, low accuracy, danger, and high cost. For this reason, the intelligent recognition of cracks based on digital image processing technology has gained wide attention. However, the irregularity of crack surface texture as well as noise affects the recognition accuracy. To realize the intelligent recognition of cracks in common solid materials, a lightweight semantic segmentation network model BiSeNetV2 was proposed for automatic crack detection, while a crack dataset was constructed autonomously. Experiments showed that the recognition accuracy of the optimized crack recognition model was improved by 7.6%. The BiSeNetV2-based crack identification model can detect and recognize cracks properly, as well as solve many challenges associated with manual detection.