Crack identification of concrete structures based on high-precision multi-level deep learning model

Cracks reflect the safety of concrete structures. Existing crack detection has the disadvantages of low efficiency and high errors, however, using image deep learning to recognize cracks avoids the disadvantage of low efficiency. The article proposes a classification model that combines the advantages of SqueezeNet and SqueezeNext, completes the migration learning of concrete structure cracks based on YOLOv8, and determines the K-means clustering segmentation method based on grayscale images by comparative analysis. Finally, the actual width of cracks is extracted based on the orthogonal skeleton line method to complete the recognition of cracks. The results show that the accuracy of the improved classification network model reaches more than 82 %, the classification accuracy of longitudinal and diagonal cracks reaches more than 96 %, the detection accuracy of YOLOv8 model reaches more than 86 %, the segmentation accuracy reaches 98 %, and the measurement error of crack width is less than 1 %.