Landsat-TM görüntülerine minnaert düzeltmesinin uygulanması: Köyceğiz örneği
Application of minnaert correction to Landsat-TM images : Köyceğiz case
- Tez No: 66793
- Danışmanlar: PROF. DR. CANKURT ÖMERCİ
- Tez Türü: Yüksek Lisans
- Konular: Jeodezi ve Fotogrametri, Geodesy and Photogrammetry
- Anahtar Kelimeler: Belirtilmemiş.
- Yıl: 1997
- Dil: Türkçe
- Üniversite: İstanbul Teknik Üniversitesi
- Enstitü: Fen Bilimleri Enstitüsü
- Ana Bilim Dalı: Jeodezi ve Coğrafi Bilgi Sistemleri Ana Bilim Dalı
- Bilim Dalı: Belirtilmemiş.
- Sayfa Sayısı: 54
Özet
Özet yok.
Özet (Çeviri)
Today remotly sensed data from the satellites is the most convinient way to monitor the earth. However, to examine the earth in detail from this kind of data needs special kinds of remote sensing techniques. Satellite images are under the geometric and the radiometric impacts which can even destroy the information extracted by the calssification methods. There are several methods which are applied to the images previously in the receiving stations to avoid such negative impacts. But the local impacts should be handled by the users. Geometric impacts can briefly be explained as follows : 1) The irregular localization between the images and the earth should be arranged by transformation methods. 2) The height differences especially in the mountainous regions cause pixel displacements. This is known as“relief dispalecment effect”. Digital elevation models are necessary to avoid such irregularities. Radiometric impacts on the satellite images are as follows : 1) The relationship between the topography and the position of sun causes illumination differences. This effect is called as“topographic effect”. But it should not be confused with relief displacement effects which is also caused by the topography. There are several methods to improve this impact. The subject of this thesis is to apply Minnaert correction to Landsat-TM images to decrease the effect of different illuminations on the earth. 2) The path taken by the sun radiation is through the atmosphere which has the scattering and absobtion properties. Such an impact is known as“atmospheric effect”on the images. 3)The pixel groups belonging to different classes make transitions between each other. As described previously, topographic effects on the radiometry of Landsat-TM images are the result of different illuminations. Such illumination differneces are caused by the position of sun and the topography. Whileas the position of sun is described by its height and azimuth, parameters of topography is slope and aspect. That is why, the most convinient way to correct the images is to use the digital elevation models which means the model of illumination. The result of topographic of effects are the blurs or“shadows”as most known which especially occur between the opposite sides of the hills in the west-east direction. However, topogarphy does not always cause negative impacts on the images because it should not be forgatten that images were also used to examine the jeology or the topography itself. On the other hand, for the three dimensional visualisation of satellite images topography helps constructing the realistic approach to the images. Such topogarphic effects occur in mountainous regions where are also generaly covered by forests. So there are several studies which examine the impact of topography on the classification results of forests. What makes special the Landsat-TM images in this context is the resolution and its passing period which is also related with the imaging geomertry and the vuposition of the sun. First of all, the necessary DEM should have a better resolution than Landsat-TM image resolution. (4 times better than 30x30m Lansat-TM resolution, as some literature says). The other point is that, Landsat-TM passes in the morning when sun radiation is not sufficient to illuminate the earth. Various literature says that 55° of sun incidence angle is the limit to create insufficient illumination. There are several methods to avoid or better to say, to decrease the impact of topograph : 1) The easiest way to improve the classification results is to make sub-classes which include different levels of illumination classes, like“forest under shadow”class and“forest under sun”class which is not a systematic method. 2) Ratio images are the other option to decrease both the atmospheric and topogarphic effects. But the disadvantage is that, different surfaces can have the same band-ratios. 3) Topographic correction functions assumes the terrain cover as Lambertian or Non-Lambertian reflector and then model the surface reflectance. So sun incidence angle is calculated in the first step from the sun position and the topographic parameters. Such methods are systematical and but more complicated in comparison with the previous methods. Earth surface is under the impact of direct, diffuse and scattered sun light. And the surfaces are seperated as perfect surfaces, mixed surfaces and Lambertian surfaces according to their reflectivity properties. Lambertian surfaces reflects the light equally to all directions that such surfaces seem bright from all viewing angles. Lambertian models do not include the diffuse sky radiance. Non-Lambertian model takes diffuse and scaterred radiance from neighbourg surfaces into consideration. In addition, Non-Lambertian models uses the same surface parameters with Lambertian Model like sun incidence angle but also includes various constants which represents surface reflectivity properties. Minnaert correction is derived from Non-Lambertian models. So it considers Minnaert constant which represents each surface reflectivity. Minnaert constant is dependent to surface roughness, image band and surface type, which is extracted by regression analysis for each surface type on the image. The parameters which belong to Minnaert correction are as follows : 1) Sun position which means sun height and azimuth. 2) Topography which is described by means of slope and aspect. 3) Minnaert constant extracted by regression analysis. Sun position and topography parameters are the elements to extract sun incidence angle which is the angle between the surface normal and the sun ray. The following formula gives the cosine of the sun incidence angle /. cos ı = cos e cos z + sın e sm z cos e = Surface Slope z = Solar Zenith Angle «l>,-) VUl(h = Solar Azimuth Angle (h = Surface Aspect The second step is to extract the Minnaert constant. Minnaert constant is the slope of the linear regression line which is formed by linearizing the Minnaert correction formula as in the following. Lnor is the normalized radiance value of the pixel, Lobs is the observed radiance value of the pixel, k is the Minnaert constant. T (Lobscose)“°r (cos/) (cose) l0g(i>ofo COS e) = 1°S(L nor) + k l0S(C0S ' COS e) Minnaert constant k is obtained by preparing the above logaritmic formula in the linear form of y = b + mx. ”k“ is the slope of the regression line. In the second step, after k is found for each band, Minnaert correction as written above ( jT, ) is performed to each pixel of the image. Study area is in the Dalaman River Basin where is in the south-west of Turkey and surrounded by one map sheet of 1: 25000 scale. Mean height in the region is 560m and the height difference lies between 100 and 1264 m which is quite great to create topographic effects. Climate is semi-arid and the region is covered by various forest types. The application was carried out by ARC-Info GIS software which is also powerful with its image processing capability. The Landsat-TM data belonged to 179. path and 34. row and its date was May- 1993. Image was rectified with 1 pixel rms. Three field works were computed previously in June- 1994-95 and 96. Position of sun was extracted from header of the data. Slope and the aspect of the region was obtained from DEM of the region. Slopes with 0° were masked as no-data because they would not be used in the next steps. After the preparation of the input parameters, the first step was to calculate the cosine of the solar incidence angle (cos(i)). Minus values of cos(i) which gave the pixels under shadow is also masked sothat they would not destroy the statistics. It should be noted that the pixels which are assumed to be under shadow can not be normalized because they do not receive any sun direct sun light. Each step is concluded by checking the minimum and maximum values of the bands which abled to understand the basic form of the illumination model in the region. The second main step was to calculate Minnaert constants for each band. After the parameters of the linear regression form are achived, regression analysis were held. The study area contained three basic vegetation cover as needled forest, mixed forest shrub and steppe. Concequently, test areas were supposed to give four Minnaert constants which were corresponding to each cover type and also to each band. However, some constrains on the field work prevented convinient correlation between the series of measurements as variables (x,y) of the regression form. IXTable 1. Specifications of The Needled Forest Test Polygons Tablo 2. The Results of Regression Analaysis for the Minnaert Correction As it is given above, only needled forest gave a sufficient correlation in the regreesion analysis which was infact nearly 0.5. The following tables give the numeric specifications of the test areas and the Minnaert constants which were derived from the needled forest test polygons generalized for the whole images. As the last step, after the necessary parameters for the each band were derived, Minnaert function was applied which gave the resultant normalized Landsat-TM image. The original and the normalized Landsat-TM images were compared by three aspects as briefly described below. a) It was observed that normalized image seemed flat in comparison with the original as visually approached. b) The difference image as three bands was obtained by substracting the normalized image from the original. The difference image which had the form of the aspect image gave the idea about which regions changed most after the correction procedure. It was band 5 which was changed most in its reflactance values, because the difference between the means of the original and the normalized image was 5 while the others were 2 and 0 respectively as band 4 and band 3. c) Classification by Isoclust clustering which is special for ARC-Info software, was applied to produce two classified images for some comparisons. The procedures had the same parameters like sample interval, number of iterations vs. 10 classes including a class Which contains both water surfaces and the insufficiently illuminated regions, are compared between each other. However, there was no chance as aerialpercentage of this class named ”insufficiently illuminated". Moreover, the classification accuracy did not improve. In this study, Landsat-TM and digital elevation data are integrated to each other to apply Minnaert correction. In this way, different aspects of topography on the images are observed. As most literature says, the behaviour of Landsat-TM data as a function of incidence angle is class-dependent. Thus, it is not suprising that a parameterization based on several cover types combined should lead no overall improvement. It should be noted that semi-empirical slope-aspect corrections like Minnaert correction in forestry applications must be class specific in order to succeed at improving classification accuracies. As some advices for proceeding studies : Regression analysis are dependent to test areas which are registered during the field works. So test areas should be distrubuted to the region according to their aspect and slopes properties and should have homoginety. Besides this, application of Minnaert correction which takes a long time with its hierarchical steps, needs an automatic program which is not pocessed in most GIS packages. XI
Benzer Tezler
- Orman varlığının saptanmasında uzaktan algılama verileri
Remote sensing data for determining forest resources
MUHİTTİN İNAN
Doktora
Türkçe
2004
Ormancılık ve Orman Mühendisliğiİstanbul ÜniversitesiOrman Mühendisliği Ana Bilim Dalı
PROF.DR. KADİR ERDİN
- Spektral indekslerin arazi örtüsü/kullanımı sınıflandırmasına etkisi: İstanbul, Beylikdüzü ilçesi, arazi kullanımı değişimi
Effect of spectral indices over land use/cover classification: İstanbul, Beylikduzu district, land use change
ÖZGE KAYMAN
Yüksek Lisans
Türkçe
2015
Jeodezi ve Fotogrametriİstanbul Teknik ÜniversitesiGeomatik Mühendisliği Ana Bilim Dalı
PROF. DR. AYŞE FİLİZ SUNAR
- Ankara metropolitan alanının yeşil alan sisteminin analizi
Analysis of the green area system of ankara metropolitian
AZAT YEŞİL
Yüksek Lisans
Türkçe
2006
Peyzaj MimarlığıYıldız Teknik ÜniversitesiŞehir ve Bölge Planlama Ana Bilim Dalı
PROF. DR. SEMRA ATABAY
- Uydu görüntülerinin piksel ve nesne tabanlı sınıflandırma sonuçlarının karşılaştırılması (Doğu Trakya bölgesi örneği)
Comparison of the pixel-based and object-based classification results of satellite imageries (Case of Eastern Thrace)
SERDAR BAYBURT
Yüksek Lisans
Türkçe
2009
Jeodezi ve Fotogrametriİstanbul Teknik ÜniversitesiJeodezi ve Fotogrametri Mühendisliği Ana Bilim Dalı
PROF. DR. DERYA MAKTAV
- Coğrafi bilgi sistemleri ve uzaktan algılama teknikleri yardımıyla toprak erozyonunun belirlenmesi üzerine araştırmalar
Başlık çevirisi yok
MEHMET ALİ ÇULLU