Urban Road Extraction Of High-Resolution Remote Sensing Image Based On Frequency Domain Filtering

For a long time,remote sensing image feature extraction mainly adopts a pixel-based method.This method extracts or classifies features based on the spectral features of the feature,that is,the pixels of the same spectral feature are clustered in the same feature area.However,even with high-resolution remote sensing images,it is impossible to avoid the situation of “homologous spectrum” and “foreign matter with the same spectrum”,and even more so.From the perspective of roads,for roads with shadows,covered by street trees,and covered by tall buildings,even high spatial resolution remote sensing images cannot basically be identified and classified by spectral characteristics.Therefore,ground objects recognition has gradually begun to change from "pixel" to "object characteristics".The paper is based on Parseval's law of conservation of energy,combines with the technical means related to spectrum energy analysis,considers and filters various characteristics of roads,and analyzes and extracts roads in IKONOS high-resolution remote sensing images within the study area.The main research results include:? Analyzing the radial and angular distributions of the spectrum diagrams of different road conditions based on IDL,it is proved that the directional characteristics of the road objects have a strong response in the spectrum diagram.This feature can be used to enhance the road linear texture,and it can also be used as one of the features to extract road from most other objects.? Based on the results of spectrum analysis,three different filters are used to filter the image by the direction of the road.Compared with the filtering results of several typical images filtered by the kernel convolution filter,Gabor and Log-Gabor filter after parameter tuning,it was found that the Log-Gabor filter exhibited better road linear texture extraction results.? The watershed segmentation method is used to segment the texture extraction results of Log-Gabor filter,and the road information is extracted from the segmentation object using the road's extensibility and directionality characteristics.Finally,the qualitative analysis and quantitative analysis are used to analyze the extraction accuracy.The results show that the method of this paper has good effect on the extraction of straight urban roads,and has poor extraction effect on the conditions of curved roads and multiple roads.Generally speaking,it can extract most of roads and form road networks,which can meet the precision needs of human daily applications.In this paper,based on the spectral characteristics of the road in high-resolution remote sensing images,the filter with appropriate parameters is used to linearly filter the image in a given direction,and the road texture information is extracted.Finally,the road's extensibility and directional characteristics are used to extract road objects.Accuracy evaluation of the extraction results.The idea of using the combination of spectrum features and road features to extract roads used in this paper has certain reference value for the extraction of other specific features.