Research On Bridge Disease Identification And Classification Based On Image Deep Learning

Bridge safety has always been concerned.Whether there is any problem in the bridge is usually assessed by the most direct expression apparent condition before determining whether to do further inspection.Therefore,it is an effective means to ensure the normal operation of the bridge by paying close attention to the apparent structural state of the bridge,timely discovering and dealing with its apparent diseases.In recent years,it is a rising research focus by using deep learning to detect a large number of bridge apparent pictures and identify the diseases’ type and location,whose related research achievements are still few.Because of the complexity of the scene and the coexistence of many kinds of diseases,there are numerous challenges in the disease identification.Aiming at the complex scenes of the regular detection of bridge apparent diseases,we mainly used deep learning algorithms combined with image processing methods to carry out research.We got a bridge apparent disease identification classifier called BADIC and developed a web application tool based on it.For the image detection project of bridge apparent diseases,the main innovation points and research achievements in this thesis are:1.A bridge apparent disease detection model based on Faster R-CNN was proposed and evaluated over the data set Bridgedisease₂019.The results show that by improving the initial setting and selection mechanism of anchor Windows,the average precision of bridge disease is increased by 3.3%and 2.3%respectively.2.For the complex detection scenes,the identification classifier BADIC was designed to realize the function of classifying,recognizing four kinds of bridge apparent diseases and extracting cracks.The recognition accuracy of the classifier for honeycomb pitting,salt out,crack and rust are 94.28%,94.71%,94.13%and 97.71%,respectively.3.The Web application tool for bridge apparent disease detection was developed,which can be applied in bridge detection engineering.The module of bridge apparent disease classification and recognition and the module of bridge apparent crack extraction were developed based on BADIC.The identification classifier BADIC and Web tools developed in this thesis meet the engineering requirements of bridge detection with strong practicability,and are effectively applied in complex shooting scenes and intricate disease scenes.