Algorithm Research And Engineering Implementation Based On Bridge Crack Automatic Identification Technology

Bridge safety is one of the most important parts of public transportation safety.Regular inspection of bridges is the most effective evaluation method for bridge safety.The standardization and accuracy of inspections are directly related to national economy and people's livelihood.At present,the regular inspection of highway bridges is mainly manual.When inspecting bridges,bridge inspectors need to judge the type of bridge diseases first,and then make positioning,cause and qualitative judgments of various diseases,and on this basis Bridge disease concludes.Due to the complexity of the inspection site environment and the need for various judgments by the inspection personnel,the inspection personnel have high requirements for their business level and their workload is quite complicated.In order to solve this situation,the thesis first analyzes,captures and forms the disease feature and cause rule database of bridge crack disease characteristics,then proceeds from the disease pictures taken at the inspection site,and proposes an automatic recognition convolution model based on bridge crack disease pictures(BCDP-AR),in-depth study of the cracks of the bridge disease,and according to the disease characteristics and cause rule base comparison,to achieve the effect of automatic disease identification and cause prediction,so that the inspection personnel can quickly and accurately make the inspection site Judge and draw reasonable conclusions and suggestions.The research of this subject mainly solves the key problems of automatic identification and identification of bridge cracks,provides great convenience for the regular inspection of bridges,has good practical value and promotion prospects,and has achieved great success in specific engineering applications.sure.The specific research contents of this article are as follows:(1)Based on the convolutional neural network(Convolutional Neural Networks,CNN),this thesis constructs a convolution model of BCDP-AR.In the model construction,ReLU and LRN are used to constrain the objective function,which can not only improve the sparsity of the model data and prevent the overfitting of the model,but also improve the accuracy prediction of the model.Through the comparison with the corresponding paper model and the experiment on the specific engineering application data set,it shows that the accuracy of the model and the sparseness of the data have great advantages and engineering effects.(2)In the process of BCDP-AR convolution model training,the predicted values and tags are obtained by the forward propagation algorithm firstly;and then the objective function is used to quantify the difference between the predicted value output by the model and the tag value given by the training data;finally,the back-propagation algorithm is used to update the model parameters according to the difference.In the process of back propagation,the SGD optimization algorithm is used to find the direction of the optimal solution for model training,which greatly increases the model's convergence and improve the recognition effect of crack disease image.(3)Due to the lack of data sets in actual industrial applications,this article will extract crack images of hundreds of bridges from the engineering application system as training objects,and contribute part of the original crack images and the processed training and test set images as Research in engineering applications.(4)In order to make the BCDP-AR convolution model more robust,this thesis put the trained model into specific engineering applications,and the practical engineering application of the regularly inspected bridges is carried out in conjunction with the “Highway Bridge Fixed Inspection System” and its renderings are presented.