Home Articles Change detection based on Remote Sensing information model and its application on...

Change detection based on Remote Sensing information model and its application on coastal line of Yellow River delta

XiaoMei Y ang
Dr., Earth Observation Research Center, NASDA
1-9-9 Roppongi, Minato-ku, Tokyo, 106-0032, China
E-mail: [email protected]

RongQing Lan QiHe Yang
Zhengzhou Institute of Surveying and Mapping
Zhengzhou City, Henan Province, 450052, China
E-mail:[email protected]

Abstract
Information about change is necessary for updating land cover maps and the management of natural resources. Many researches have been undertaken to develop methods of obtaining change information. Based on the summarization of the methods on change information extracted from remotely sensed data, the paper promotes the method of change detection based remote sensing information model. This method is applied to detect the coastal line change of Yellow River Delta (YRD). It lays the foundation for research on the change relation of natural and human activity impact each other, and finally aids to study the regional geographic feature through more than 10 years remote sensing images in YRD.

1. Introduction
Information about change is necessary for updating land cover maps and the management of natural resources. The information may be obtained by visiting sites on the ground and/ or extracting it from remotely sensed data. For many of the physical and cultural features on the landscape there are optimal time periods during which these features may best be observed. Remotely sensed data acquired at the fixed time interval becomes an important factor. Many researches have been undertaken to develop methods of obtaining change information. Change detected from different temporal images usually reflect natural and human activity impact each other and then can be used to study how to form the regional geographic feature.

As we know, the Yellow River Delta is a delta grows fastest in the world. Because of its new land forming and unstable environment, its development is far more backward than other famous large river delta. However, the Yellow River Delta has a good geographic location, rich natural resources and tremendous developing potentiality. Its development is of great importance to the development of North China.

This paper first summarizes the methods on change information extracted from remotely sensed data. Then based on the different object models the method of change detection is mainly discussed. Finally through more than 10 years change result of Yellow River Delta, we analyze the coastal line change.

2 Change Detection Methods Based Remote Sensing Data

2.1 General Methods
Temporal feature is important and special in all the system characteristics. Because spatial and spectral information can be seen from images, temporal feature is relatively abstract, it is difficult to reflect directly. Its feature only can be seen by the change of spatial and spectral feature. Some commonly used change detection algorithms are summarized and analyzed as the following table.

Table 1 Main algorithms of change detection

Algorithm and example Method procedure Problem
Image Transformation (e.g. PCA).Fung et al., 1987, 1988.(Eastman et al., 1993, Bauer et al.,1994) Composition of different temporal data, then classified or using PCA to transform all together No classified change information
Image Arithmetic Change Detection.Price et al., 1992. Arithmetic operation among different bands, e.g. Dijk=Bvijk(1)-Bvijk(2)+C No classified change information
Post-Classification Comparison (Rutchey et al., 1994) Classification of remote sensing images, then comparison pixel by pixel Relay on classification accuracy and multiple classification
Data sources aided change detection(Lunetta et al., 1991. Map data or GIS aid to analyze Relay on quality of aided information
Spectral vector change (Michalek et al., 1993, Johnson et al., 1998) Comparison with spectral vector of different temporal CMpixel= [Bvijk(2)-Bvijk(1)]2 Suitable for region of spectral change greatly

2.2 Based on Remote Sensing Information Model Change Detection
Remote sensing information is a complicated information. It is the comprehensive behavior from a certain environment. The behaviors of images varied largely with different ground features due to their different radiation and scattering characters to visual light, infrared and microwave. As a result, we can not build their remote sensing information models for ground feature under a unified mode. Three levels of model are conducted to describe the ground feature as follows:

  1. Spectrum vector based remote sensing information model for ground feature:
    Some features, such as water body, vegetation, scene of forest fire, possess distinctive spectrum characteristic of multiple bands. Remote sensing information is often corresponding to ground object. This model is the starting point for a study. It is called primitive class model.
  2. Multi-source information based remote sensing information model for ground feature:
    Most of ground features can not rely on just a single information model. Especially when coming across the appearance of different objects but equally spectra or equally spectra but different objects, it is necessary to use multi-temporal remote sensing information or other supplementary data (supported with GIS data base) to build its corresponding remote sensing information model for ground features. Based on primitive class model, this model is created by fresh supplying new element terms according to derivate mechanism.
  3. Geo-knowledge based remote sensing information model for ground feature:
    Some ground features or appearances, especially the recessive information, can be recognized only through complicated processing and deeply analyzing. This kind of remote sensing information model for ground features not only includes spectrum vector characteristics of ground features but also needs to add geo-knowledge and expert knowledge and experience as well as the process of operating the knowledge and experience. For the reason, the study to this model deals with wide range of fields and is very complicated.

For main objects such as water, vegetation and non-vegetable, they can be extracted only by establishment of spectral feature model. For example,

Figure 1 Hierarchy of ground objects

Figure 2 Spectral curve of different objects
Water and land
With the increase of bands, the spectral reflectance value of water body decrease, i.e. bij1>bij2>bij3>bij4>bij 5>bij7.Meanwhile, band 5,7 can be used to select threshold to segment water. But due to the special effection of bedload in YRD, CH3, CH4 >CH2. So, we first using water spectral feature model to extract water, then within the region of water body, classification is conducted to obtain different depth classification map (shown as figure 3). It simply and distinctly reflects the water region distribution of YRD.

Figure 3 Water classification map
Vegetation and non-vegetation
Vegetation has high reflectance value in band 4.and in band 3 there is low reflectance. For this character, bijv=bij4/bij3 is often taken as the index to identify vegetation region and non-vegetation region. Certainly, there is much other vegetation indexes. When it is used in image segmentation, we define the following rules:
bijv>a1; vegetation region
Pixel(i,j) bijv