Research On Rapid Identification Method And Modeling Technology Of Engineering Structure Surface Cracks Under Complex Background

In engineering structures such as buildings and bridges,the presence of cracks has a significant negative impact on the durability,aesthetics,and maintainability of the structures.Current crack detection technologies gradually use machine vision with images or videos as carriers to replace human eyes for crack recognition or quantitative detection.However,accurate crack identification remains a highly challenging task,especially in complex backgrounds or in the presence of numerous interferences.Moreover,researchers mainly focus on crack detection from the perspective of image processing,with relatively less research on the visualization,digitization,and modeling of crack information.Rapid and accurate identification of crack images in complex backgrounds and their mapping to three-dimensional models for visualization,digitization,and modeling is of significant engineering significance and research value for the scientific and accurate assessment of the current health status of engineering structures.This paper focuses on surface cracks of structures and conducts in-depth research on intelligent recognition algorithms for structural cracks in complex backgrounds.Based on the crack detection results and quantification program,combined with the three-dimensional model of the actual structure and pixel space mapping theory,the paper achieves the modeling of surface damage information in the structure.To address the challenges of difficult crack detection and low identification accuracy in complex backgrounds,this paper proposes a crack multi-scale segmentation method that can automatically realize reasoning and information feedback fusion functions.By combining crack image multi-scale information with a mask inference module,the proposed method can effectively simulate human perception,reasoning,and judgment behaviors.Fully considering the slender morphological characteristics of structural cracks,and starting from the concept of multi-granularity observation information complementarity,this paper proposes a deep integration algorithm for accurate detection of slender cracks in complex backgrounds,using CNN coarse-grained classification and Transformer long-range correlation finegrained segmentation.The two crack segmentation methods proposed in the paper decompose a complex crack identification problem into multiple simple tasks,reducing the training difficulty of the detection network and improving the interpretability of the entire neural network.The performance is verified by combining measured crack images,achieving better detection results than a single model.In order to achieve the digital quantitative mapping of multi-scene crack image information to a 3D model,this paper utilizes smartphones and drones to acquire surface images of engineering structures.Based on the high degree of freedom,portability,and ease of use of smartphones,this paper combines feature point matching algorithms and pixel space mapping theory to study the morphological transformation process of cracks from world coordinate space to pixel coordinate plane,the quantitative description of crack features,and the positional relationships of cracks among multi-scale images.For surface crack images that cannot be obtained by smartphones,a drone with a built-in high-definition matrix shooting function is employed to perform shooting tasks,and a feature point matching-based image stitching method is proposed to achieve fast identification and quantitative detection of outdoor scene cracks.To realize the modeling of 2D images into 3D models,this paper maps actual structural surface cracks to 3D models in the form of digital images,establishing a transmission link for synchronous updating between actual structural damage and digital models,thus monitoring and evaluating the current health status of practical engineering structures.By utilizing smartphones and drones to obtain image data and combining crack detection algorithms,image fusion technology,feature point matching algorithms,and image mapping theory,this paper realizes the conversion process of real structural cracks from world coordinate system to pixel coordinate system,and then from pixel coordinate plane to 3D model,providing digital and modelized damage information for evaluating structural damage status and analyzing the causes of damage.