Research On Damage Identification Of Frame Building After Earthquake Based On Convolution Neural Network

The first task in the emergency period after the earthquake is to evaluate the damage degree of the building structure in the disaster area,and quickly organize the emergency risk removal,restoration and reconstruction according to the assessment results.Traditional evaluation methods are often limited by the professional background of the appraisers,which makes the evaluation results vary from person to person,and the evaluation process is long and timeconsuming.Therefore,a fast and stable damage assessment method is particularly important.Based on the analysis of the research status at home and abroad,this paper proposes a method of building structure damage assessment based on convolution neural network,which has high recognition efficiency and accuracy.The main research contents are as follows:(1)In view of the large number of RC frame structures in China,the failure pictures of frame structures in Wenchuan earthquake and Lushan earthquake are collected,including non structural members and structural members.According to the standard for classification of damage levels of buildings,the beams and columns of structural members are divided into slight damage,medium damage and severe damage;the ceilings and infilled walls of non structural members are divided into intact,slight damage,moderate damage and severe damage;for the problem of insufficient pictures,the image library is expanded by means of rotation and mirror image to provide data set for subsequent model training.(2)In this paper,based on the migration learning method,the Alex Net network in the deep convolution neural network framework is used to learn the damage data set of the post earthquake frame building.Through the contrast experiment with Google Net and Res Net,the superiority of Alex net in this recognition task is verified.This paper studies the influence of batch size,epoch,dropout,optimization function and other parameters on the recognition results,and puts forward a ceiling failure recognition model with an accuracy of about 96%;aiming at the problem of low accuracy in the recognition of the damage degree of infilled wall and the damage degree of beam and column,an improved convolution neural network based on Alex Net network is proposed,which improves the recognition accuracy by about 4%.(3)According to the standard of building damage classification,the quantitative evaluation criterion of building structure damage grade is put forward.The damage identification results of non-structural components and structural components are integrated by MATLAB,and the damage degree of building structure caused by earthquake is obtained.In order to facilitate operation,GUI graphic interactive interface is also developed.