|
|
|
Geology Disaster 2
|
An automatic technique and its effect by using Landsat TM data to extract hydrothermal alteration information in subtropical volcanic rock area
Zhao Tuanhong, Zhang Fuxianf, Chen Nanfeng
Remote Sensing Division, Dept. of Earth Sciences,
Zhejiang University
Hangzhou, China
Abstract
A new automatic technique for using Landsat TM data to extract hydrothermal alteration zones n subtropical volcanic rock area covered by vegetation is developed here, and its applications is very successful in Bamao Au and Ag multimetal mineral area, Xinchang country, Eastern Zhejiang province. Not only are known alteration zones in the area displayed effectively, but also several unknown hydrothermal alteration zone, silicified quarts viens, ore-controlling structures and volcanic apparatus, which are very useful for prospecting, are revealed here.
Introduction
Landsat TM and 7(2.08-2.35 mm), in hydroxyl-bearing clay minerals exhibit strong absorption (2.20 mm) 1,3,4, is very useful for geological exploration. With the developing of Landsat-4,5, a series of studies for mapping altered rocks have been done by many schlars 4,5,6,7, and have got some valuable results. But the recent literatures show that there are still two problems in this field:
- Most people select arid or semiarid area as study are. The effective studies of technique and application are still lacked under range vegetation, especially in the subtropical area.
- Enhancing the alteration information, people generally delineate the altered rocks by visual interpretation instead of automatic technique. Such a case with a bad accuracy and low working efficiency, cannot meet the needs of geological exploration.
Therefore, it is in an interested problem how to use Landsat TM data for altered rocks enhancement and automatic extraction in vegetated area.
Alteration Zone Automatic Extraction
For improving the application effects, a new effective technique, which can use Landsat TM data for mapping alteration zone automatically under range vegetation, is designed here; its flow chart is show in Fig 1.
Its working principle as following :
- Alteration Information Enhancement
TM5/7 is considered as an effective band ratio for alteration information enhancement in arid or semiarid area. But in vegetated area, vegetation impedes the alteration information discrimination of TM5/7 ratio image, because it is both widely distributed in the surficial environment and can be spectrally similar to clays when sampled by TM8,6. Analysis of the study area show that both green vegetation and clay minerals contribute to the lighter pixels on the TM5/7 image, and most of vegetation has a larger TM5/7 ratio than clays. Whereas we must find a new method to remove the effects of vegetation, so that we can enhance the alteration information effectively in such a area covered by vegetation.
It is well known that TM4/3 band ratio serves as an excellent vegetation index image. The analysis of the study area shows that patches of vegetation correspond to the lightest pixels and far lighter that others do on the TM4/3 ratio image.
Fig. 1. Flow chart of alteration zone automatic extraction
The principle component transformation is used for analyzing the TM5/7 and TM4/3 band ratio image here. Once computed, the first principal component (pc#1) contain the contribution of main information in both ratios-that is vegetation; the second principal component (pc#2) contains the contribution of altered minerals. Such a method is called as "directed principal component analysis (DPCA)" by S.J Fraser and A.A Green. They have used this method for anlaysing ATM data in tropical savannah woodland with 50to 70 percent total vegetation cover, and the clay and vegetation have been identified successfully, but few people have done such a study with TM data in subtropical area.
This method has been developed for using Landsat TM data enhance the alteration information here, and its application is very successful in Bamao area.
- Alteration zone automatic Extraction
On the PC#2 image, the lightest pixels correspond to altered rocks, and the lightness is vegetally increased with altered rocks. For extracting the alteration zones exactly, following steps are taken here:
- Density slicing. A bilateral selection method, combining both sample training of known alteration zones and histogram training of PC#2 is used to select a threshold that can describe the real digital number (DN) range of altered pixels.
Supposed that DN range of altered pixels is (D_e, 255), where D+e is lower limit, D is an approximate lower limit, and e is a correction factor (e<=0).
An approximate mean value (M) of known altered pixels is calculated by statistics of alteration zone. Considering that the M is often decreased due to other alteration free pixels inside the "Statistics window", and relatively speaking, weaker alteration zone has little significance for geological exploration, we let D=M. The histogram of PC#2 can display the distribution sate of DN over all images directly. According to the signature of histogram near D, a best "e" can be found to determine the threshold, so that we can slice off the alteration information. Letting altered pixels be 1, other be 0, a(0,1) binary image (ALTA) is acquired.
- Random noise removing. On the ALTA image, random noise, coming from TM3, 4,5,7 band data, is one disturber. For removing such a disturbance, we have made further analysis to determine the statistical signatures of "1" pixels on raw TM data, and a "discrimination window" of alteration information is selected. Then such "1" pixels, being outside the "window" and being considered as random noise, can e removed, an a new image (ALTB) is made here.
- Alteration zone delineating. By further analysis of ALTB image, we find that there are two types of information displayed on it, one is contributed by altered rocks with small scale and weak alteration, which is almost no significance for geological exploration, and another is contribute by the altered rocks with, large scale and strong alteration, which had important significance for geological exploration. We must find a method to delete the first one and remain the second one, but alteration zone is displayed as discrete cluster on the image, some other information, such as water bodies, residential points inside the alteration zone, have disturbed it. So it is difficult to filter such information using ordinary methods.
A new filter algorithm - called as contexture filter, is presented here: a first, we use a "close" algorithms of "Dilation-erosion" to connect all these discrete "1" pixels one alteration zone: and then statistical signature of "1" pixels in spectrum and space is computed with scanning and tracing method" At last, sample training is made to determine the principles for filtering, so that we can extract the valuable alteration information.
Using this algorithms, not only can the weakness of ordinary filter methods, with which discrete points inside the valuable alteration zone will be deleted, be overcome, but also the edge effects that will be induced by contexture or texture methods can be avoided. Its effects are very satisfactory.
- Displaying. Using "close' method to connect eh discrete clusters into consistent clusters; adding TM1 images a background image on it ; Letting different colors represent varieties of alteration grades respectively. Then an alteration zone displaying image, on which had topographic information, is made well.
|
|
|
|
|
|
|
