Research And Application Of Asphalt Pavement Core Sample Image Recognition Technology Based On Deep Learning

Asphalt pavement core sample reflects the real condition of pavement,so it is an important basis for asphalt pavement inspection and analysis and quality evaluation.However,at present,asphalt pavement core samples are mainly used in mechanical properties test and extraction analysis,and the properties of aggregate particles and asphalt mixture in core samples need to be further studied.Based on this,this research takes the side surface of asphalt pavement core sample as the research object,integrates computer vision,deep learning and digital image processing technology,and proposes a complete scheme integrating image acquisition,image segmentation,image processing and equipment research and development to mine the pavement information contained in asphalt concrete core sample.The main research contents and results of this research are as follows:(1)Based on the theory of 3D reconstruction and the principle of image mosaic synthesis,a RMAIS registration model-assisted image synthesis method suitable for asphalt pavement core sample is proposed.An area scan camera was used to take images of the core side surface to reconstruct the 3D point cloud model.The standard model was registered with the point cloud model,and the accurate mapping relationship between the expanded plane and the core surface was established.The ultra-clear distortion removal expansion diagram of the core side surface was synthesized by image cutting and stitching.(2)The semantic segmentation neural network U-Net ++ is used to segment aggregate particle components in asphalt concrete images.In this research,a labeling acceleration method was proposed to intelligentally capture aggregate particle contour,the data set was expanded by sliding clipping,the segmentation result was optimized by Gaussian fuzzy algorithm,and the training model was optimized by using quadratic labeling active learning method.(3)The segmentation results of each mixture grade(SMA-13,AC-20,AC-25,SUP-20)were evaluated.Threshold segmentation methods such as Otsu were used to obtain binarization images,and then the segmentation results of aggregate particles were performed with holes filling and adhesion segmentation.A slider pixel analysis model was established to realize the segmentation of the void components and asphalt mortar components.(4)Based on the results of image segmentation,the slenderness ratio and roundness morphological characteristics of aggregate particles are analyzed,and the calculation model and defect analysis and judgment method of asphalt concrete overall uniformity are proposed.A set of asphalt pavement core analyzer is developed according to the research content of this research,and the software and hardware design and engineering application are realized.