The application of regional factors in the extraction of Remote Sensing information

The application of regional factors in the extraction of Remote Sensing information

Chen Minzhen
Lab. of Agricultural Remote Sensing
Zhejiang Agricultural University

Zhao Yuanhong, Chen Lan, He Kan
Lab. of Remote Sensing University,
Zhejiang Remote Sensing Centre China

Abstract
This paper selects Junshan Hill Area, which lies between Dongting Lake and Yangtze River, as an experimental area, and a TM image (flood season) as experimental data. On the basic of studying spectral features and distribution patterns of various surface target, the experimental area is divided into three types: the field inside embankments, the transitional field outside embankments, and the water field. Some types of surface target, which have identical or similar spectral features but lie in different regional factors, can be distinguished according to regional factors. Other surface target which have different spectral features but lie in the same regions, can be distinguished by making use f threshold extraction method. Several different type of field are successfully distinguished in this way, they are the reed field, the reed weed field, the mud sand river water containing high mud sand content, lake water containing mid mud-sand content, the low mud-sand content water inside embankments, and the flooded area. The research indicates that it is feasible and effective to introduce regional factors into the extraction of RS thematic information. The approach is conducive to the efficient monitoring of flood.

Introduction
Te application of remote sensing information greatly depends on the efficient extraction of various thematic information. Nowadays, computer automatic extraction (classification) has became the mainstream of research on information extraction techniques. Because of the restriction of spectral and spatial resolution in RS information, researchers introduce some other sources of information in order to extract RS thematic information greatly. This paper presents and approach of how to extract various types of water information efficiently and accurately from RS images, and discusses a general method of how to introduce some regional factors into the extraction of RS thematic information according tot eh distribution patterns of natural surface target. Junshan hill farm is selected as a experimental area which is lies between Yangtze River and Dongting Lake. A TM image (Aug. 25 1987 flood season) is used as the experimental data.

The extraction of thematic information from TM image
The basic information of various surface targets contained in TM image is the spectral information, i.e. the brightness on each band of TM image. The main surface targets in this experimental area are paddy field, arid field, various type of water, settlement, reed field, reed-weed field, and mud-sand beach. The brightness of each surface target is shown in Fig. 1. On the basic of analyzing he brightness, this paper puts forward a multi-level threshold extraction method. The method takes into full account various types of water and easily flooded reeds, both the which have strong interest in the information system of flood danger forecast (called "system" below for short), the approach includes several steps:

Fig.1 Brightness diagram of surface target in TM image

Extraction of water information
The basic requirement of the system is to extract various types of water efficiently and accurately. As shown in fig.1 the spectral difference between water and other surface targets is relatively obvious. By analyzing the histogram and comparing the extracted results. The researcher indicates that the ratio RM1/(TM3+TM7) combined with threshold method can extract water information more efficiently. The result is shown Fig. 2. The red are expresses water field extracted by threshold method. The white area shows the difference between the water field extracted respectively by the cluster classification method and by threshold, method. As compared with the boundary given by the cluster classification, the boundary drawn by threshold method coincides more accurately with the real condition. In order that different types of water can be further distinguished, a fig. 2 Water area extracted from, ratio image TM5/TM4 is from TM image produced for it has a high sensitivity to mud sand content in water. On the principle of multi level classification, water can be divided into several types by means of threshold method, including: the river water containing low mud-sand content, and the lake water containing low mud-sand content, and the lake water of which mud sand content in quantity is between the two types of water mentioned above. The classification tree is shown in Fig 3 Fig. 4 gives the classification results. The results indicate that all types of water have been accurately classified, especially the river water, the main parts of flood, having much more satisfying accuracy of classification.

Fig.2 Water area extracted from TM image

Extraction of flooded Area
Investigation into flooded area is one of the jobs the system has to do, in flooded are, water flows slowly and the mud sand content in water is decreased. The spectral features of that kind of water is similar to that of water enclosed by embankments (see Fig. 1). Thus the water in flooded area is classified as a type of water containing low mud sand content. In order to make a distinction between the flooded area outside embankments and the water are inside embankment the area is divided into three types according to the distribution characteristics of regional surface targets. The three types are the area inside embankments, water area and the transitional reap outside embankments. The distribution patterns of surface target in this area is shown in table 1. The flooded area outside embankment and the water area inside embankment are alike in spectrum, but the two types of areas le at different regions according to the distribution patterns. Therefore they can be easily distinguished. The classification tree is shown in Fig 5. the blue area in Fig 4 represents the scope of flooded area.

Table 1. The distribution patterns of natural surface targets in the Research areas.

The area inside embankments	The transitional area outside embankments	Water area
Arid field, paddy field, settlement water, rod, dike garden field.	Reed field, Red-Weed field, mud-sand beach	River water lake water (in dry season)

Fig.3 The Classification Tree

Fig.4 The Results Derived Multi-Level Threshold Method

Extraction of Reed information
The reed, one of economic paints, has a wide range of distribution in the area being rich in river and lake. Also reed is one of surface target that can be easily flooded in flood season. As shown in fig 1 the reed spectral characteristics is nearly identical to that of arid field or paddy field. In other words, these types cannot be accurately divided only by making use of TM spectral information. On the basic of distribution patterns (see table 1), the reed field can be easily extracted because the reed grows in the transitional area outside embankments, while and field or paddy field is located inside embankment. Some types of surface target such as reed field, reed-weed field, and mud sand beach, are all distributed in the transitional area. In order that these different types of surface target lying at the same region can be distinguished, a ratio image TM4/TM3 is produced which is sensitive to the vegetable index. On the principle of multi - level classification, the boundary of different types of area can be drawn by means of threshold method (see gfig.5). The result is shows in fig4. As comparing with the result given by conventional method, he boundary of reed field and reed-weed field coinside with the real boundary precisely. To sum up, some surface targets, which have same spectral features and different distribution patterns, can be accurately distinguished according tot eh principle of multi level classification. On each level, a simple threshold method which takes into account regional factors is used in order to extract various thematic information quickly and accurately. This approach is flexible, adaptable, and easy to be realized. Fig 5 indicates a programme of multi level classification. Fig 4 shows the results achieved on the programme in fig 5.

Fig.5 Multi-Level Classification Tree

Conclusions

Water Information can be quickly extracted from the ratio image TM1/(TM5+tM7) by means of simple threshold method.
On the basis of multi level classification, various types of water containing difference mud-sand content can be efficiently distinguished when ratio image TM5/TM4 is dealt with. The result, especially for the water containing high mud sand content in flood season, are very satisfying.
Some surface target have similar spectrum lie at different region. In such circumstances, regional factors should be introduced in order to extract RS information more accurately. Such a method of thematic information extraction is suitable not only to TM images but also to other sources of RS information. Almost all surface target have their regional distribution patterns, so it is positive that introducing regional factors into the matic extraction of RS information has a general significance.
A ratio image TM4/TM3 is sensitive to the vegetable index. From TM4/TM5 image different types of reed fields can be successfully distinguished.
On he principle of multi-level classification, many classification trees be flexibly constructed by means of different combinations of regional factors with spectral characteristics and many types of thematic information ca be extracted to meet different requirements, such a method is flexible, adaptable, and simple in computation. It is ideal approach for the thematic extraction of RS information.

References

M.Goldberg, Region-Based Modeling Algorithms for Remotely-Sensed Data, Machine processing of Remotely Sensed data symposium, 1984.
Slephen R.Yool, Performance of image processing algorithms for classification of Natural Vegetation in The Mountains of Southern California, Int. J. Remote Sensing, 1986 Vol. 7, No.5.
B.V. Dasarathy, A Composite Classifier system, Design: Concept and Methodology, proceed of the IEEE, 1979 V. 67, No 5.