Recognition And Diagnosis Of Structural Surface Cracks Based On Machine Vision Technology

The identification and diagnosis of cracks on the surface of the structure is an important part of assessing the safety and durability of the constructions.Timely mastering the length,width and direction of cracks plays a vital role in maintaining the health status of structures and putting forward reasonable treatment schemes.The traditional crack detection method is time-consuming,low in detection accuracy and detection efficiency,and there are many factors that affect the evaluation of results.Therefore,it is necessary to introduce an objective and efficient non-contact fracture recognition and diagnosis method.With the development of imaging and photography technology,people have begun to apply machine vision technology to process crack images to replace manual closecontact identification and diagnosis.Aiming at the disadvantages of the traditional manual detection method and combining with the image processing research of machine vision technology,this paper proposes a crack recognition algorithm of steel box girder bridge and a crack cause diagnosis method of shield lining image respectively.The main research contents of this paper are as follows:(1)In order to deal with the crack segmentation task of super-resolution steel box girder bridge image with complex background,this research proposes an ensemble deep neural network with three sub-networks contains human observation,reasoning and decision-making processes.First,crack detection classifier and crack semantic segmentation network in the subnetworks of the ensemble network are used to simulate multi-scale human visual processes,and then an innovative crack inference classifier based on the crack morphology features of the surrounding area is proposed to strengthen or weaken the information about the existence of cracks.The prediction probability fusion of image and inference information is used to judge and correct the crack repeatedly.The results show that the proposed algorithm can effectively reduce the interference from handwriting and welded joints,and provide more accurate crack segmentation results for the measurement of crack characteristic parameters.(2)For the surface crack image of subway shield tunnel,an image matching method is proposed to deduce the crack cause mechanism of tunnel segment under the condition of unclear damage information and incomplete structural parameters by identifying and marking cracks.By simulating a variety of working conditions that may cause the segment of shield damage through finite element,and then converting the three-dimensional segment data model into two-dimensional images to establish a case database,the real pipe cracks are matched with the case database according to the location and development form of the cracks,finally the preliminary diagnosis of the causes of shield segment cracks is realized.