Decomposition apparatus and method for refining composition of mixed pixels in remote sensing images

Patent metadata

Field	Value
Publication number	US-9317929-B2
Application number	US-201313915101-A
Country	US
Kind code	B2
Filing date	Jun 11, 2013
Priority date	Jun 14, 2012
Publication date	Apr 19, 2016
Grant date	Apr 19, 2016

Abstract

To achieve space correlation of pixel decomposition results and reduce noise problem caused by isolation, there is provided a decomposition apparatus and method for refining composition of mixed pixels in remote sensing images, including: a preprocessing step for temporarily determining the provability value of the composition ratio of the different land cover types of each pixel in the image, based on a received remote sensing image and spectral information, to obtain first material composition information; and a neighborhood correlation calculation step for analyzing the correlation between a main pixel and the neighboring areas by using the first material composition information of each of the pixels present in the neighboring areas within a predetermined range in which the pixels are adjacent to each other, and optimizing the first material composition information of the main pixel by the result of the correlation analysis, to obtain second material composition information.

First claim

What is claimed is:

1. A decomposition method for refining composition of mixed pixels in remote sensing images to decompose the composition of different land cover types of a pixel in which a plurality of types of land cover are mixed, by using the spectral refection characteristics of the different land cover types, wherein the decomposition method comprises:
a receiving step for receiving a remote sensing image and spectral information corresponding to the received remote sensing image;
preprocessing step for temporarily determining a probability value of a composition ratio of different land cover types of each pixel in the image, based on the received remote sensing image and the spectral information, to obtain first material composition information; and
a neighborhood correlation calculation step for analyzing the correlation between a main pixel and the neighboring areas of the main pixel, by using the first material composition information of each of the pixels present in the neighboring areas within a predetermined range in which the pixels are adjacent to each other, and optimizing the first material composition information of the main pixel by using the result of the correlation analysis, to obtain second material composition information,
wherein the preprocessing step includes performing preprocessing for the image to be decomposed, extracting training samples corresponding to the different land cover types present in the actual area with respect to the image to be decomposed, and calculating the probability value of the composition ratio of each land cover type in each pixel by using the statistical value of the extracted training samples, to obtain the first material composition information.

2. The decomposition method for refining composition of mixed pixels according to claim 1 ,
wherein the neighborhood correlation calculation step analyzes the correlation between the main pixel and the neighboring pixels by the Markov Random Field.

3. The decomposition method for refining composition of mixed pixels according to claim 1 ,
wherein the neighboring areas of each pixel are the neighboring areas on the two-dimensional plane.

4. A decomposition method for refining composition of mixed pixels in remote sensing images to decompose the composition of different land cover types of a pixel in which a plurality of types of land cover are mixed, by using the spectral refection characteristics of the different land cover types, wherein the decomposition method comprises:
a receiving step for receiving a remote sensing image and spectral information corresponding to the received remote sensing image;
preprocessing step for temporarily determining a probability value of a composition ratio of different land cover types of each pixel in the image, based on the received remote sensing image and the spectral information, to obtain first material composition information; and
a neighborhood correlation calculation step for analyzing the correlation between a main pixel and the neighboring areas of the main pixel, by using the first material composition information of each of the pixels present in the neighboring areas within a predetermined range in which the pixels are adjacent to each other, and optimizing the first material composition information of the main pixel by using the result of the correlation analysis, to obtain second material composition information,
wherein the neighboring areas of each pixel are the neighboring areas on the three-dimensional plane,
wherein the neighborhood correlation calculation step analyzes the correlation between a pixel to be analyzed and the neighboring areas of the pixel in the two-dimensional space, and
wherein the neighborhood correlation calculation step analyzes again the correlation with the pixel to be determined as the parent pixel, by controlling the neighborhood region used for the decomposition of the pixel to have a certain stretch based on the image resolution and the correlation analysis result, by determining the new neighborhood region with the stretch as the upper area present on the third dimension, and by determining the second composition information of the parent pixel as the first material composition information of the corresponding pixel in the upper area, in order to achieve correlation analysis on the multi-dimensional space.

5. The decomposition method for refining composition of mixed pixels according to claim 4 ,
wherein the neighborhood correlation calculation step determines the stretch of the neighborhood region by the multi-scale image pyramid method.

6. A decomposition method for refining composition of mixed pixels in remote sensing images to decompose the composition of different land cover types of a pixel in which a plurality of types of land cover are mixed, by using the spectral refection characteristics of the different land cover types, wherein the decomposition method comprises:
a receiving step for receiving a remote sensing image and spectral information corresponding to the received remote sensing image;
preprocessing step for temporarily determining a probability value of a composition ratio of different land cover types of each pixel in the image, based on the received remote sensing image and the spectral information, to obtain first material composition information; and
a neighborhood correlation calculation step for analyzing the correlation between a main pixel and the neighboring areas of the main pixel, by using the first material composition information of each of the pixels present in the neighboring areas within a predetermined range in which the pixels are adjacent to each other, and optimizing the first material composition information of the main pixel by using the result of the correlation analysis, to obtain second material composition information,
wherein the neighborhood correlation calculation step performs the correlation analysis by repeating the neighborhood correlation calculation several times, and gradually optimizing the first material composition information of the image region to obtain the second material composition information.

7. A decomposition method for refining composition of mixed pixels in remote sensing images to decompose the composition of different land cover types of a pixel in which a plurality of types of land cover are mixed, by using the spectral refection characteristics of the different land cover types, wherein the decomposition method comprises:
a receiving step for receiving a remote sensing image and spectral information corresponding to the received remote sensing image;
preprocessing step for temporarily determining a probability value of a composition ratio of different land cover types of each pixel in the image, based on the received remote sensing image and the spectral information, to obtain first material composition information; and
a neighborhood correlation calculation step for analyzing the correlation between a main pixel and the neighboring areas of the main pixel, by using the first material composition information of each of the pixels present in the neighboring areas within a predetermined range in which the pixels are adjacent to each other, and optimizing the first material composition information of the main pixel by using the result of the correlation analysis, to obtain second material composition information,
wherein the neighborhood correlation calculation step sets the neighborhood region to an 8-pixel neighborhood region.

Assignees

• Hitachi Ltd

Inventors

• Zhang Dai
• Zhang Xue

Classifications

• G06T7/11 Primary

  Region-based segmentation · CPC title

• G06T7/404

  Physics · mapped topic

• G06T7/0081

  Physics · mapped topic

• G06T2207/20076

  Probabilistic image processing · CPC title

• G06T7/0087

  Physics · mapped topic