Image Recognition Of Weathering Bridge Steel Corrosion Based On Deep Neural Network

With the rapid development of the economy,China's bridge construction level has been at the forefront of the world,and the requirements of bridge construction on steel are getting higher and higher.In addition to requiring steel to have higher rigidity,strength,and good weldability,many projects also require that the steel has good corrosion resistance.Weathering steel is favored by increasing number of bridge designers because of its more outstanding corrosion resistance than traditional carbon steel and the lower cost in the entire bridge life cycle.However,the detection of the rust state of the rust layer during the use of weathering steel bridges has not been able to be well solved.The rust state of the rust layer on the surface of weathering steel bridges is often determined by manual visual inspection that is time consuming and prone to make mistakes,which also leads to a longer period of manual inspection and miss the best maintenance opportunity for weathering steel bridges.Therefore,it is of great significance to study the evaluation method of the corrosion state of weathering steel.Based on this purpose,this article has done the following research.In this paper,the corrosion law of weathering steel was studied by simulating the corrosion in industrial atmospheric and marine atmospheric environments,and compared it with the corrosion law of ordinary carbon steel under the same environmental conditions.On this basis,we further studied industrial The effect of chloride ion and sulfate ion in the industrial atmospheric and marine atmospheric environment on the stabilization of the rust layer of weathering steel were further studied.While studying the corrosion laws of steel,pictures of weathering steel corrosion samples and ordinary carbon steel corrosion samples within different corrosion environments and different corrosion cycles were obtained to obtain weathering steel corrosion image data sets and ordinary carbon steel corrosion image data sets.On the basis of the experiment,the weather recognition steel corrosion image data set was used for preliminary training and testing of the image recognition algorithm based on deep neural network.The results of the initial training and testing are far from the expected accuracy.On this basis,the image recognition algorithm is improved.Using the weathering steel corrosion image data set obtained in this experiment to train and test the improved image recognition algorithm based on deep neural network,the results show that the training accuracy of image recognition can reach 91%,and the testing accuracy of image recognition can reach 84%.Then the ordinary carbon steel data set obtained by the experiment was used to train and test the improved algorithm.The results show that the training accuracy and test accuracy are basically stabilized above 94%.The reason why the accuracy of image recognition is higher than that of weathering steel is because that the size of the base metal of ordinary carbon is more regular,and under the same corrosion environment conditions,the characteristic difference between the corrosion images of each cycle of the ordinary carbon steel is more obvious.Finally,the improved algorithm is used to identify the concrete cracks.The results show that the training and testing accuracy are quickly stabilized at more than 95% during the identification process,because the difference between the characteristics of the concrete image with and without cracks is very obvious,and there are sufficient concrete crack image data set samples.