Automatic identification and time series evolution analysis method of concrete cracks based on computer vision

Traditional manual detection of concrete cracks is inefficient and subjective, and it is difficult to realize continuous monitoring. This paper proposes an automatic identification and time series evolution analysis method of concrete cracks based on computer vision. This method constructs an end-to-end intelligent monitoring system, innovatively integrates attention-oriented double-flow network, graph neural network (GNN) and fracture mechanics theory, and realizes static accurate identification of cracks, inter-temporal case correlation and evolution trend prediction. Firstly, a double-stream network of space-channel joint attention mechanism is designed, which combines local details and global context features to effectively suppress complex background interference and improve the accuracy of micro-cracks identification. Secondly, GNN is introduced to model the fracture examples in continuous time series images, so as to solve the complex evolution problems such as fracture and merger of cracks and realize the "life course" tracking of cracks; Finally, the Paris fracture mechanics model and LSTM (Long Short-Term Memory) data-driven method are combined to construct a hybrid time series model driven by mechanism and data, so as to improve the prediction accuracy and physical consistency of crack width and length evolution. The experiment was carried out on the basis of 15,000+ images of three bridges in China for two years. The results show that the F1 score and mIoU score of this method are 94.6% and 89.8% respectively, which is significantly better than the mainstream methods. The accuracy of fracture instance association is 96.2%; The root mean square error (RMSE) for crack width prediction was reduced to 0.015 mm, with a mean absolute percentage error (MAPE) of 7.3%, demonstrating excellent long-term prediction stability. This study provides an efficient, intelligent, and sustainable technical approach for concrete structure health monitoring, holding significant engineering application value.