The Research Of Road Information Enhancement And Extraction Method On Remote Sensing Image

Road target identification and attribute inversion in both civil and military have important practical value. Because main materials of roads are asphalt, cement, sand and dirt, roads in the remote sensing spectrum are different from other spectral features of ground objects. At the same time, in remote sensing image road shows space characteristics of the directional extension as a straight line or curve. At present, the identification and extraction of road methods are mainly the edge detecting method, region partition method, template extracted method and classification technology of object-oriented method, angular texture signature method and method based on vector machine(SVM) method, etc, but these methods exist problems such as the existence of mixed pixels, extraction accuracy is not high.This paper mainly studies building measured spectral characteristics identification index to roads, space enhancement method,decomposition of mixed pixels. First, for the study of spectral characteristics, ASD FieldSpec 4 field measured hyperspectral data, then the material such as asphalt, cement and soil of the three kinds of roadbeing the same category, separatelycalculatewith water, soil and vegetation, with variance analysis method to select the optimal band building roads index, discussing its extraction conditions and applicable scope. Second, space characteristics of the study, using existing differential operator respectively extracts the edge of the road, choosing the optimal operator and setting threshold combined with mathematical morphology to complete the extraction of spatial information. To make up for the inadequacy of road indexextraction, application based on the Minimum Noise Fraction of Mixture Tuned Match Filtering(MNF- MTMF) model to carry on the decomposition of mixed pixels in the measured spectra of the endmember in landsat8 OLI multispectral data to extract the road information, finally according to the spectral characteristics and spatial characteristics of road information enhancement and extraction and calculating accuracy.The following conclusions is obtained by experiment in this paper:1. Using variance analysis selects landsat8 OLI data NIR band and GREEN band as the optimal band to build Ratio Road Index(RRI),Difference Road Index(DRI) and Normalized Difference Road Index(NDRI). NDRI recognition effect is best, but the extraction precision is not high, only the width that is larger than or close to width of a road pixel can be extracted, the rest specifications of the road extract incompletely, so it has some limitations.2.The result of using the Canny algorithm on the extraction of road space information is better than using the Sobel algorithm, Prewitt algorithm, Log algorithm, which can get more complete road edge information.3. In the aspect of the integrated spectral information and spatial information, because the road pixels of the OLI data on road are mixed pixels mostly, using the MNF-MTMF model of decomposition of mixed pixels to enhance and extract is good to solve the problem of using the NDRI, which can get more integrity road information.Comprehensive combination of MNF- MTMF model and Canny algorithm to extract the road information has good effect, the edge of the road is clear, the internal pixels of the road are full. Extraction result accuracy: pixel misclassification rate is about 49.2%, and the rate of pixel leakage points is about17.6%.