| Earthquakes are one of the natural disasters,which seriously threaten social and economic development and the safety of people’s lives and property.The evaluation of the earthquake damage level of the building structure after the earthquake is one of the main tasks of post-disaster reconstruction.The degree of cracking and fracture of the reinforced concrete column is the main basis for the evaluation of the earthquake damage level after the earthquake.At present,it is mainly through experienced experts to observe the degree of cracks and fractures of reinforced concrete columns to evaluate the post-earthquake damage level.The workload is heavy and the human eyes are easy to fatigue.Some subtle cracks cannot be observed,and a comprehensive scientific evaluation cannot be given.Aiming at the existing problems,this paper uses twin neural network to establish a post-earthquake damage evaluation model for reinforced concrete columns.On the one hand,the twin neural network is used to determine the damage level,and on the other hand,small cracks are observed through the thermal map,and the earthquake is measured from both macro and micro aspects.The damage grade of the reinforced concrete column is evaluated after the earthquake,which provides a scientific reference and judgment basis for the on-site workers after the earthquake.First,analyze the evaluation methods of post-earthquake damage,collect test images in the laboratory of Engineering Mechanics Research Institute through earthquake simulation experiments,and calibrate the damage levels and establish standards based on the corresponding relationship between the seismic damage level and the force of the reinforced concrete column.Collect real seismic damage images of reinforced concrete columns through literature search and on-site photography,and calibrate the damage level according to the judgment of experts.The reinforced concrete column in the image is extracted as image data,and the seismic damage image data set of the reinforced concrete column is constructed.The experimental image data is used as the training set and the verification set,and the real earthquake damage image data is used as the test set to prepare for neural network training and testing.Secondly,build a post-earthquake damage assessment model of reinforced concrete columns based on siamese network.The metric learning method and attention mechanism are added to the twin neural network model,and the network model is optimized and used for post-earthquake damage assessment of reinforced concrete columns to improve the accuracy of the damage grade judgment of reinforced concrete columns.At the same time,gradient-weighted activation mapping（Grad-CAM）technology is incorporated into the post-earthquake damage model of reinforced concrete to generate a Grad-CAM heat map,which can display small cracks in reinforced concrete columns and solve the problem of damage positioning of reinforced concrete columns.Finally,the post-earthquake damage assessment model of reinforced concrete columns is applied to the real earthquake damage images of reinforced concrete columns for damage assessment,the damage level after the earthquake is determined and the surface damage heat map of reinforced concrete columns is generated,and the human-computer interaction interface is designed through Py Qt5 to verify this The effectiveness of the model in evaluating seismic damage levels of reinforced concrete columns in practical applications. |
