Urban road recognition method based on multi-spectral images

Abstract

The invention provides an urban road recognition method based on multi-spectral images. The method includes: 1, based on a object-oriented segmenting method, segmenting a road and surrounding ground objects in a Urban road recognition; 2, extracting low-level features of each segmented area, establishing mapping rules from low-level features to high-level semantic objects, realizing mapping from low-level features to high-level semantic features of an image, and constructing a semantic model for road recognition, wherein the low-level features include geometric features and spectral features,and the high-level semantic objects include green belts, lane lines, and road potential areas. The method adopts the method of combining the SLIC super pixel and the structure tensor rough segmentation, has a good anti-noise ability, and can obtain better segmentation results for complex urban multi-spectral images. The road recognition method based on semantic knowledge solves the problems of lowroad recognition accuracy, tendency of producing holes and fractures in the complex environment of multi-spectral images.

Description

technical field [0001] The invention belongs to the technical field of image recognition, and in particular relates to a multispectral image urban road recognition method. Background technique [0002] Traffic road target recognition in remote sensing images is one of the key technologies in the theory of automatic target recognition. Urban roads are an important part of geographic information databases. How to accurately identify urban road targets from multispectral images is of great significance for surface detection, urban structure description, and road renewal. [0003] The background of multi-spectral images is complex, and it is easily affected by the shadow of surrounding objects and the spectral approximation of roads and houses, so it is difficult to guarantee the recognition accuracy. For the urban road recognition of remote sensing images, the existing technologies mainly include road extraction methods based on pixel level and road extraction algorithms based...

AI Technical Summary

Problems solved by technology

[0004] Due to factors such as vehicles and pedestrians, the road binary image extracted by the pixel-level road extraction algorithm is prone to "salt and pepper phenomenon", and the center line extracted on this basis is prone to breakage

The complex texture and context features of the multispectral image itself, as well as the lack of universality of the existing segmentation algorithm itself, make the road extraction algorithm based on the object hierarchy prone to adhesion phenomenon

Images