نگاشت مناطق آلتراسيون با استفاده از مدل آميخته ي گاوسي و نقشه بردار زاويه طيفي

Mapping Alteration Zones using Gaussian Mixture Model and Spectral Angle Mapper

در ايران باتوجه به وسعت زياد كشور و گستردگي مناطق پتانسيل ­دار ذخاير معدني، شناسايي و مديريت صحيح ذخاير اهميت بسيار بالايي دارد. از طرف ديگر به دليل پيشرفت­ هاي فن­آوري سنجش از دور و توليد داده­ هاي فراطيفي با اطلاعات طيفي فراوان، استفاده از اين داده ­ها جهت مطالعه­ ي دقيق پديده­ ها به سرعت در حال گسترش است. تصاوير فراطيفي به دليل نمايش گسترده خصوصيات طيفي عوارض و پديده ­هاي سطح زمين در بسياري از علوم مرتبط با زمين مورد توجه قرار گرفته اند. از آن­جايي كه بسياري از مواد معدني در محدوده مادون قرمز كوتاه ويژگي­هاي طيفي منحصر به­فردي دارند، تشخيص و تفكيك آن­ها به كمك تصاوير فراطيفي به جهت پهناي باندي بسيار كمي كه دارند، بسيار حائز اهميت است. در اين پژوهش قصد داريم به تفكيك مناطق آلتراسيون مرتبط با مس پورفيري به كمك تصاوير فراطيفي هايپريون و با استفاده از دو الگوريتم نقشه بردار زاويه طيفي (SAM) و طبقه ­بندي كننده مدل آميخته گاوسي بپردازيم. شايان ذكر است كه انجام پيش ­پردازش ­هاي لازم بر روي تصاوير فراطيفي هايپريون به دليل داشتن نسبت سيگنال به نويز پايين، يكي از گام­ هاي بسيار ضروري به جهت رسيدن به نتايج مطلوب مي­باشد. به منظور مقايسه و ارزيابي صحت روش­هاي پيشنهادي در اين تحقيق، ابتدا داده تصويري از سنجنده هايپريون مورد استفاده قرار گرفت. همچنين يك داده­ شبيه ­سازي شده مورد استفاده قرار گرفت، تا كارايي الگوريتم ­هاي مورد استفاده در اين تحقيق را نشان دهند. براي ارزيابي نتايج نيز از ماتريس خطا به كمك داده ­هاي زميني استفاده شد و مولفه ­ها­ي دقت كلي و ضريب كاپا به عنوان معيار ارزيابي مورد استفاده قرار گرفتند. دقت كلي و ضريب كاپا براي الگوريتم SAM و آميخته گاوسي به ترتيب معادل با 82% ، 0/75 و 80% ، 0/71 حاصل شد.

Due to the extent of mineral deposits, identification and proper management of resources is very important. According to the advent of remote sensing and specially producing hyperspectral remote sensing data which can get abundant spectral information, using this data for detailed study is rapidly expanding. Launch of the EO-1 in November 2000 introduced hyperspectral sensing of the earth from space through the Hyperion system. Hyperion has a single telescope and two spectrometers in visible near-infrared (VNIR) and short-wave infrared (SWIR). These spectral bands could provide abundant information about many important earth-surface minerals. Therefore one of the main aim of the present study was to examine the feasibility of the EO-1 Hyperion data in discriminating and mapping alteration zones around porphyry copper deposits (PCDs). The study area is situated at the Central Iranian Volcano-Sedimentary Complex, where the large copper deposits like Sarcheshmeh as well as numerous occurrences of copper exist. The visible near infrared and shortwave infrared (VNIR-SWIR) bands of data were used for image classifying and specially alteration mapping. The Pre-processing which was implemented on the level 1R Hyperion data in order to remove noise and acquire surface reflectance includes five steps that named removing uncalibrated bands, spatial displacement correction, destriping, spectral curvature (smile) correction and at last atmospheric correction. It is noticeable that atmospheric correction, because of using the target detection algorithm, SAM, is one of the most important step in this study. Therefore the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) algorithm, available in ENVI software, was implemented to obtain surface reflectance data. This algorithm is a MODTRAN4-based atmospheric correction software package designed to eliminate atmospheric effects through derivation of atmospheric properties such as surface albedo, surface altitude, water vapor column, aerosol, and cloud optical depths, as well as surface and atmospheric temperatures from hyperspectral data. In this paper Spectral Angle Mapper (SAM) and Gaussian Mixture Model (GMM) were implemented on pre-processed and calibrated Hyperion dataset. For using SAM algorithm, introducing reference spectra is obligatory. Information extraction from a Hyperion data set involves several processes including extraction of scene spectral endmembers using an integration of MNF, pixel purity index (PPI), and n-dimensional visualizer approaches. Then the extracted spectra which characterized using spectral analysis procedure available at ENVI and visual inspection, were used as reference for subsequent processing by SAM algorithm. On the other hand Gaussian mixture model (GMM) has been successfully used for HSI classification. It has also proved beneficial for a variety of classification tasks, such as speech and speaker recognition, clustering, etc. For estimating the parameters of GMM, the Expectation-Maximization (EM) algorithm was used. In order to compare and assess the accuracy of methods proposed in this study, a simulated data used to demonstrate the efficiency of algorithms which used in this study. Results revealed that Hyperion data prove to be powerful in discriminating and mapping various types of alteration zones while the data were subjected to adequate pre-processing. Overall accuracy and kappa coefficient for results of SAM and GMM are 82%, 0.75 and 80%, 0.71 respectively.