| In recent years,remote sensing and its image processing technology have developed rapidly.With the continuous improvement of the resolution of remote sensing images,many research results have been achieved for road extraction.Road materials(such as asphalt roads,cement roads,and dirt roads)are an important parameter of road attributes,and the urgent need and importance of information extraction are becoming more prominent.The information of road materials in high-resolution remote sensing images has the characteristics of diversity,complexity and uncertainty.For example,irregular changes in the material of low-level rural roads,road areas blocked by trees or buildings,etc.These accurate extraction of road material information increases interference and difficulty.Aiming at the above problems,this paper uses high-resolution remote sensing images as data sources to extract multi-source features of road materials.Take advantage of the powerful nonlinear statistics and classification capabilities of support vector machines(SVM).This paper realizes the extraction of road material information by multisource feature fusion.The main research contents and results are as follows:(1)High-resolution remote sensing image preprocessing.Experimental area data was obtained through orthorectification,radiation correction and geometric correction.Obtain the road surface information of the road through the manual marking method and screen suitable samples and test samples.On the basis of road extraction,the arc segmentation method is used to extract the secondary information of the road.(2)Multi-source features information extraction algorithm for road materials.Using the secondary information extraction of the road,the local HSV color feature extraction algorithm is improved,and the color information of the road material is extracted while avoiding the influence of surrounding features.The rotating LBP operator is improved.The average value of the 3×3 area is used as the new threshold.And the average value of the rotated LBP value is used as the final LBP value to extract the texture features of the road material.It improves the operator's noise resistance.(3)Road material information extraction based on multi-source features.Perform dimensionality reduction and normalization on multi-source feature data,and retain the part that contributes more than 90%.Then it adds the GLCM feature to supplement the loss information and fuse through the fusion strategy to obtain the two-by-two fusion feature and the three-fusion feature.This paper builds a grid search SVM model and genetic algorithm SVM model.The best model selected by cross-validation is the genetic algorithm SVM model,and the best feature fusion method is a combination of three feature fusion methods,which realizes the extraction of road materials.Aiming at the problem of shorter arcs caused by secondary segmentation of arcs,this paper proposes an object-based road connectivity optimization algorithm and an arc-based road connectivity optimization algorithm: 1.By connecting arcs connected with shorter arcs to form new arcs,using the new arcs for multi-source feature extraction and material identification.2.Using statistical methods,set the type of material with the highest vote in the new arc segment to the new arc segment material.Both methods can improve the problem of short-arc road misidentification to a certain extent,optimize road connectivity characteristics,and improve the accuracy of road material identification.(4)Software module design and implementation.A functional module for road material information extraction of remote sensing images is designed.This module contains algorithms for secondary extraction,feature extraction,and fusion of road thinning arc information.It can quickly and easily extract road material information of remote sensing images. |
PDF Full Text Request