Bridge Crack Morphological Classification Using GAN And Deep Transfer-learning

Concrete cracks is a direct representation of building health condition.Methodologically bridge cracks can be mainly divided into mainly four type: horizontal cracks,vertical cracks,reflective cracks,and mesh cracks.The different forms of cracks usually indicate different bridge health problems.Failure to deal with it or ignorance would definitely cause damage on the supporting capacity of the bridge and shorten the service time of the bridge.In some cases,it will lead to major safety accidents.Therefore,it is necessary to accurately recognize the morphological characteristics of bridge cracks.Bridge crack images are different from traditional rock crack images or pavement crack images.It has many complex characteristics,such as a variety of noises,diverse and complex background textures,irregular distribution,and uneven lighting.In view of the traditional bridge crack classification method,we find that the complex background and noise would largely affects the classification results.In this paper,we proposed a new method that combines GAN and deep transfer learning.This paper conducts research based on image data of bridge cracks collected by drones.The main research contents of this article are as follows:(1)This paper discussed the major causes of cracks,the complex characteristics of crack images and the morphological characteristics of cracks.Secondly,we introduced the establishment of bridge crack database.(2)We compared the traditional FT saliency detection algorithm and the improved saliency optimization detection algorithm based on conditional generative adversarial network(CGAN).Compared to traditional method,our CGAN based proposal can extract high-quality cracks structure from its complex background.(3)To evaluate our proposed,we conducted comparative studies with prevalent deep transfer learning models Inception-v3,Inception-resnet-v2,Xception and Densenet 201.An end-to-end bridge crack morphological classification and recognition algorithm integrated with deep transfer learning.The experimental results show that the algorithm for classifying and identifying the bridge crack morphology based on the conditional generation adversarial network designed in this paper can avoid the interference of complex characteristics on the bridge crackclassification and identification.At the same time,it can also be applied to small data sets,improve the convergence speed,control over-fitting,reduce training time,improve the generalization ability of the model,effectively extract the crack interest areas,and improve the efficiency and accuracy of bridge crack classification and recognition.