كاربرد ماشين بردار پشتيبان در تفكيك زون‌هاي دگرساني هيدروترمال با استفاده از سنجنده آستر

Application of Support Vector Machine Algorithm to Discriminate the Hydrothermal Alteration Zones Using ASTER Sensor

در اين پژوهش با استفاده از سنجنده آستر تلاش شده است كاربرد الگوريتم ماشين بردار پشتيبان در تفكيك دگرساني‌هاي هيدروترمال ذخاير مس پورفيري بررسي ‌شود. براي آموزش اين الگوريتم در مجموع 2204 پيكسل از مناطق كاني‌سازي شده انتخاب شد. باندهاي 4، 6، 7 و 8 سنجنده آستر براي شناسايي دگرساني‌هاي فيليك و آرژيليك و 9 باند محدودۀ مرئي و مادون قرمز نزديك براي شناسايي دگرساني پروپيليتيك به‌عنوان ورودي اين الگوريتم انتخاب شدند. به‌منظور ارزيابي خطاي طبقه‌بندي، ماتريس درهم آميختگي بررسي شد. نتايج ماتريس در هم آميختگي بيان‌گر آن است كه خطاي طبقه‌بندي براي زون فيليك و آرژيليك نسبتاً بالاست و امكان تفكيك اين دو زون به سادگي امكان‌پذير نيست در حالي‌كه دگرساني پروپيليتيك به‌خوبي طبقه‌بندي شده است. هم‌چنين اين تحقيق تابع خطاي جديدي به‌نام خطاي كور را تعريف كرد كه با استفاده از ماتريس درهم آميختگي نسبت پيكسل‌هاي دگرسان طبقه‌بندي نشده را محاسبه مي‌كند. بر اساس مقدار اين خطا، ماشين بردار پشتيبان 6/73 درصد از پيكسل‌هاي دگرسان را طبقه‌بندي نمي‌كند. هم‌چنين صحت كل طبقه‌بندي الگوريتم ماشين بردار پشتيبان برابر06/66 درصد و ضريب كاپا برابر6522/0 است

This work intends to apply ASTER images to discriminate hydrothermal alteration zones in Kerman Cenozoiic Magmatic Belt (KCMB). Band ratio, principal component analysis, Crosta and color composite images are important methods to analyze satellite images. Previous researches showed that these techniques are not able to discriminate hydrothermal alteration zones and they usually detect vegetation covering as alteration zones. The reason is found in the spectral signature of vegetation and alteration minerals. It means that they present the same interaction when face with electromagnetic energy in different wavelengths. Hydroxyl-bearing minerals are the important products of hydrothermal alteration. Clays, which contain Al-OH- and Mg-OH-bearing minerals and hydroxides in alteration zones, are distinguished by absorption bands in the 2.1–2.4 µm range of ASTER data. Solving these problems is difficult when using standard image-processing techniques such as band rationing, principal component analysis, or spectral angle mapper. In recent years, several attempts were made to extract altered regions in the areas covered with vegetation. To overcome this problem, this research uses ASTER data by applying support vector machine (SVM) algorithmn. SVM is a new technique for data classification in remote sensing application. This paper aims to investigate the potential of SVM algorithm in mapping of hydrothermally altered areas. In many applications, SVM has been shown to provide higher performance than traditional learning machines and has been introduced as powerful tools for solving classification problems. The adopted dataset contains three ASTER scenes using SWIR and VNIR bands, covering the Meiduk porphyry copper deposit, Kader, Abdar and Iju occurrences located in Kerman Province, southeast Iran. Material and methods This work has been prepared on three ASTER level 1B scenes. Two scenes were acquired on 18th April 2000 and another scenes on 15th June 2007. These scenes were georeferenced by using an orthorectified ETM + image, in UTM projection and WGS-84 ellipsoid as a datum. The first two data sets were corrected for Crosstalk. Atmospheric corrections were also performed by using Fast Line of Sight Atmospheric Analysis of Spectral Hypercubes (FLAASH). The data sets were then mosaicked.­­ Internal Average Relative Reflectance (IARR) correction was also applied. In this part, the training and test samples of the ASTER data are presented. The adopted image is a multispectral satellite image that contains 2204 training pixels which 516 pixels are related to arjillic zone, 1278 pixels are related to phyllic zone and 500 pixels are pertinent to propylitic zone (Fig. 1).