Identification Of Concrete Crack Based On Deep Learning

Concrete is one of the most widely used and largest used building materials,and it is also widely used in various industries,such as housing construction,bridges,and roads.After the concrete structure is poured,due to the common influence of internal and external factors such as low tensile strength and self-shrinkage of concrete,and environment temperature changes,cracks of different order of severity and forms will inevitably occur during the process of constructing and operating.As the crack continues to develop,once the size of the crack exceeds a certain limit,it will not only weaken the aesthetic feeling of the building,but also affect the normal use of the structure and its durability.Therefore,it is necessary to regularly detect the occurrence and propagation of cracks on the surface of concrete structures,and it is indispensable to pre-control the test results,which can effectively reduce the damage of concrete structures and their effects.To this end,this thesis starts with deep learning,deeply studies the convolutional neural networks,and further develops the method of concrete crack identification based on target detection.The research content mainly includes the following three aspects:(1)Crack identification based on CNN and sliding window algorithm.A concrete crack detection model was designed.The model can automatically learn the effective features of crack image classification on concrete surface,and realize the automatic classification of concrete crack image.Then the model was compared w:ith the current mainstream model algorithm in this paper.Finally,the sliding window technology is used to realize the crack identification and location.(2)Crack identification based on deep migration learning.In order to reduce the demand quantity of data for the crack detection model,a deep migration detection model based on VGG-16 was designed.At the same time,different fine tuning layers were compared and determined.(3)Crack identification based on YOLOv3.Applying the latest deep learning target detection method,it can quickly and accurately identify cracks on concrete surface and frame the specific location of concrete cracks in the image.The innovations of this paper include:(1)Designing a concrete crack detection model based on convolutional neural networks for concrete cracks;(2)Comparing and analyzing the VGG-16 network based on deep migration learning to determine the best fine-tuning layer,to provide the basis for the relevant personnel to choose the appropriate fine-tuning layer.(3)Applying YOLOv3 to concrete crack detection,and optimizing it.