Research On Multi-Scale Characterization And Segmentation Evaluation Method Of Fractures Based On Similarity Contrast Framework

Pavement diseases are an important factor that threatens traffic safety and affects riding quality.Using image information technology for pavement crack detection is a widely used technical method in the intelligent maintenance of transportation infrastructure.Accurately extracting the geometric features and spatial structure of pavement diseases is a key element in ensuring the long-term service performance of the pavement.However,in actual engineering,there are various types of pavement image acquisition devices,and diseases have complex and diverse sizes and shapes with strong multi-scale features.Existing image feature extraction algorithms are difficult to accurately characterize the geometric structural features of cracks,and inaccurate disease extraction results bring deviation to subsequent analysis.How to improve the accuracy of pavement disease extraction and establish an objective segmentation evaluation index is an urgent problem to be solved in practical applications such as pavement disease detection.This study focuses on asphalt pavement crack images and uses techniques such as local binary patterns,contrastive learning,and manifold learning to finely extract the geometric structural features of cracks.By understanding the similarity relationships between various pavement disease samples,the quality of crack segmentation methods can be objectively evaluated without human intervention.In this paper,a multi-scale crack structure characterization method based on multi-radius local binary patterns is proposed for the complex structural features of cracks.The method focuses on the edge information in the crack image,uses contrastive learning to create training sample pairs,and automatically assigns relative importance to each scale to accurately characterize the crack’s geometric structure.Then,various segmentation methods,including classic digital image processing techniques and deep learning network models,are used to generate multiple sets of crack binary images with different segmentation results.Finally,a similarity-based crack segmentation evaluation method is proposed to automatically select reliable crack extraction results.At the same time,in order to verify the effectiveness and universality of the algorithm,experiments were also carried out on shale and alluvial fan data sets.The experimental results show that compared to existing image feature extraction methods,the multi-radius local binary patterns method can more accurately extract the multi-scale features of cracks and quantify the morphological similarity between various types of crack structures.The similarity contrast framework can quantitatively analyze the ability of crack extraction methods to retain the pavement structure change rules and use easily quantifiable objective criteria in evaluating segmentation quality,effectively reducing human intervention.It has played a positive role in promoting the segmentation and detection of pavement cracks in practical engineering and can provide important references for highway transportation intelligent maintenance decision-making.