A high-resolution remote sensing image change detection method and system

Abstract

The invention discloses a high-resolution remote sensing image change detection method and system. The method comprises the steps that firstly, carrying out super-pixel segmentation on two high-resolution remote sensing images through a simple linear iteration algorithm SLIC, and obtaining a first super-pixel segmented image and a second super-pixel segmented image; Extracting spectral features and texture features of the first super-pixel segmented image and the second super-pixel segmented image; Calculating feature difference vectors of the first super-pixel segmented image and the second super-pixel segmented image according to the spectral features and the texture features; Obtaining a first variation amplitude image and a second variation amplitude image according to the characteristic difference vector; Fusing the first amplitude-of-change image and the second amplitude-of-change image to obtain a unified amplitude-of-change image; And obtaining a change image by using a conditional random field algorithm according to the unified change amplitude image. According to the method, generated noise can be eliminated, boundary information of a changed image can be well reserved, and the accuracy of a detection result is improved.

Description

technical field [0001] The invention relates to the field of remote sensing image recognition, in particular to a high-resolution remote sensing image change detection method and system. Background technique [0002] Change detection is a very important part of remote sensing image research. Its purpose is to analyze the differences between images acquired at different times in the same place. In recent decades, the results of change detection have been widely used in many fields, such as: environmental monitoring, disaster assessment, urban planning, etc. [0003] In multi-temporal remote sensing images, the development of change detection technology has attracted extensive attention of scholars. In the past few decades, researchers have proposed many different change detection techniques for medium and low resolution remote sensing images (such as MODIS, Landsat (MSS, TM), SPOT, etc.). unit. The pixel-based change detection method first uses two time-phase images to ob...

AI Technical Summary

Problems solved by technology

With the improvement of remote sensing image resolution, it brings more opportunities to change detection technology, but also brings many difficulties

Because the improvement of spatial resolution can extract more abundant spectrum, texture, structure and detail information, but compared with the change detection of medium and low resolution remote sensing images, the change detection of high-resolution remote sensing images is more difficult, mainly reflected in the following two aspects : On the one hand, with the improvement of spatial and spectral resolution, it is more complicated to use the information of its spectrum and texture to calculate; It is not friendly to pixel-based change detection techniques, and the results often have a lot of salt and pepper noise

It is reasonable for traditional change detection technology to be applied to medium and low-resolution remote sensing images, but it has the following four huge defects when applied to high-resolution remote sensing images: If there is a large registration error, a large number of false detections and missed detections will be caused due to the high spatial resolution of high-resolution remote sensing images, which will seriously affect the accuracy of change detection; secondly, pixel-based change detection is easily affected by shadows and seasonal changes, and will not The detection of the change class is a change class, resulting in false detection; moreover, the determination of the changed and unchanged classes by the threshold method is unstable, and it is easy to be interfered and cause false detection and missed detection; finally, due to pixel-based changes The detection technology only uses a single pixel as a processing unit, and cannot effectively use the spectral and spatial context information between images. The advantage of high-resolution remote sensing images lies in the extraction of rich detail information.

Therefore, limited by the problem of its own structure, the final result is similar to that of object-oriented change detection, and the detection accuracy cannot be significantly improved, and it is easy to produce over-smoothing at the boundary

Images