A Directed Matched Filtering Algorithm (DMF) for Discriminating Hydrothermal Alteration Zones

A Directed Matched Filtering Algorithm (DMF) for Discriminating Hydrothermal Alteration Zones using the ASTER Remote Sensing Data

Abstract

 

This study applies Minimum noise friction (MNF) transformation and pixel purity index (PPI) respectively to extract effectively endmembers on the ASTER airborne hyperspectral data to obtain the distribution map of alteration minerals in the area and uses virtual verification to verify the results. This paper also introduces “Directed matched filtering “algorithm which tries to find a purest pixel based on endmember features. Result evaluated with virtual verification and show agreement of 100%. DMF algorithm has two user defined output based on expert decision. High matching and lowest matching..

The objectives of the study included (i) extract purer pixels from more mixed pixels for determining endmembers (PPI); (ii) Matched Filtering (MF) was used to determine certain alteration for four minerals; (iii) Nonuse of threshold on  MF result (NTMF) was introduced to identification 283 target pixels and made it possible to be evaluated. The evaluation shows an agreement of 100% between the results of the MF analyses and virtual verification.

 

 

Keywords:

 

1. Introduction

Magmatic activity within the Kerman Cenozoic Magmatic Arc (KCMA) is generally related with porphyry copper and skarn mineralization, involving deposits such as Sar Chesh-meh, Sungun, Meiduk and many other economic ore bodies (Dimitrijevic, 1973; Zarasvandi et al., 2005). Hydrothermal fluid processes generate porphyry copper deposits and alter the mineralogy and chemical composition of the country rocks. Phyllic, argillic and potassic zones are produced by hydrothermal alterations in porphyry type deposits (Lowell and Guilbert, 1970). Also, an oxide zone is expanding over many of the porphyry bodies, which are rich in iron oxide minerals (Azizi et al., 2010). Specific mineral assemblages are produced by alteration with diagnostic spectral absorption features in the visible and near infrared (VNIR) through the shortwave infrared (SWIR) (0.4–2.5 μm) and/or the thermal infrared (TIR) (8.0–14.0 μm) wavelength regions (Pour and Hashim, 2011). These diagnostic spectral absorption properties in the visible and near infrared through the shortwave length infrared regions could be recognized by multispectral and hyperspectral remote sensing data as a tool for the initial stages of porphyry copper exploration (Kruse et al., 2003; Di Tommaso and Rubinstein, 2007; Gabr et al., 2010; Pour et al., 2014; Pour and Hashim, 2015). Hydroxyl-bearing minerals are the most extensive products of hydrothermal alteration. Clays and sheet silicates, which comprise 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 due to molecular vibrational processes.

High concentrations of clays and sheet silicates are specified by very high reflectance in band 4 of ASTER data. These features of phyllosilicates have been applied in remote sensing investigations for mineral exploration (e.g., Livo et al., 1993; Hubbard and Crowley, 2005; Mars and Rowan, 2006; Rowan et al., 2006; Tangestani et al., 2008). The vital task in the remote sensing exploration is to determine and discriminate alteration zones of porphyry deposits, specifically potassic and phyllic, because the main parts of Cu, Mo and Au mineralization could be found in potassic, phyllic and moderate argillic zones (Cox and Singer, 1986; Lowell and Guilbert, 1970; Evans, 1993). Discrimination between these types of alteration minerals is difficult when using some image-processing techniques such as band rationing, principal component analysis, or spectral angle mapper. The mentioned methods are not useful in distinguishing alteration zones. In recent years, several attempts were developed to identify altered regions. Abrams et al. (1983) attempted to identify hydrothermal alteration zones using digitally processed aircraft multispectral images. Kaufmann (1988) recognized hydrothermal alteration using digitally processed TM images. Knepper and Simpson (1992) applied TM color ratio composite images to map hydrothermally-altered rocks. Bennett et al. (1993) used TM data with field and laboratory data to study the alteration zones. Goosens and Kroonenberg (1994)   applied TM ratio images to discover rocks overlain by residual soil. Carranza and Hale (2002) integrated results of TM images and ground data to identify hydrothermal alteration. Porwal et al. (2004) used a neuro-fuzzy system in providing a mineral potential map from several exploratory datasets. Honarmand et al. (2011) applied principal component analysis and spectral angle mapper to recognize hydrothermal alteration minerals. Bodruddoza and Fujimitsu (2012) tried to classify alteration zones using the colour composite, band ratio, principal component, least square fitting and reference spectra analysis. Molan et al. (2014) applied the matched filtering on the HyMap airborne hyperspectral data to obtain the distribution map of alteration minerals in the Maherabad area and used virtual verification to verify the results. They introduced moving threshold technique in which an appropriate threshold value is found to specify the types of alteration zones. Matched filtering (MF) method is one of the spectral mixture analysis which performs partial unmixing (e.g.Harsanyi and Chang, 1994; Boardman et al., 1995). To determine the existence of the target in each pixel, MF is calculated by correlating the known spectrum of the target with an unknown signal. In other words, the MF algorithm calculates the cross-correlation between the noisy signal and the pure input one (Heusdens et al., 2012). This research aims to consider the outputs of MF algorithm and improve them using a spectral signature-based filter. In this type of classification, called directed matched filtering (DMF), the image analyst could supervise the pixel categorization process by specifying numerical description of the various spectral features of reference spectra (endmember). The DMF procedure requires considerable interaction with the analyst, who must direct the classification by identifying the spectral features of the target spectra. In other words, the DMF uses the spectral characteristics of the target to discriminate other pixels in the image with similar characteristics.

2. Geological setting

The study area is part of the Central Iranian Cenozoic Magmatic Belt (CICMB), which runs parallel to the Zagros geo-suture for about 1800 km from Azerbaijan Province of Iran in the northwest to north of Makran in the southeast (Fig. 1).

The area is located in the Kerman Cenozoic magmatic arc (KCMA), which is part of the southeastern sector of the CICMB.

The oldest rocks in the study area are Cretaceous coloured mélange that consists of sedimentary and volcanic rocks; the youngest rocks are Quaternary alluvial deposits and gravel fans (Hassanzadeh, 1993). Argillization, sericitization and propylitization are the most common types of hydrothermal alteration in the area. The Dehaj and Aj phases of volcanic activity, which produced pyroclastic, dacitic and basaltic rocks, occurred in the Pliocene (Soheili, 1981). The study area contains copper mineralization at Meiduk, Abdar, Kader, Godekolvari, Iju, Serenu, Chahfiroozeh and Parkam (Dimitrijevic, 1973).

Fig. 1: Geological map of the study area (modified after Soheili, 1981)

 

3. Data set

The ASTER is an airborne imaging instrument onboard the terra platform, the flagship satellite of NASA’s Earth Observing System AM-1 (EOS AM-1), which was launched in December 1999. The instrument consists of three separate subsystems of the electromagnetic spectrum and resolution; the visible and near-infrared (VNIR) sensor consists of 4 bands; short-wave infrared (SWIR) consists of 6 bands and thermal infrared (TIR) sensor consists of 5 bands (Abrams, 2000; Abrams et al., 2004). This work used three ASTER scenes previously pre-processed up to Level 1B. Two scenes were acquired on 18 April 2000 and one scene was acquired on 15 June 2007. These scenes were georeferenced using an orthorectified enhanced thematic mapper plus (ETM⁺) image, in the UTM projection (zone 40) with the WGS-84 ellipsoid as a datum. The first two scenes were corrected for Crosstalk. Internal Average Relative Reflectance (IARR) correction was also applied on the images. The data sets were then mosaicked.

4. Reference spectra extraction

To apply spectral analyzing algorithm, it is needed to select reference spectrum (or “endmember”). There are two ways to select the reference spectrum, the first: it could be selected from spectral libraries, the second: the reference spectrum could be extracted from the image. The reference spectra extracted from the latest method are often more useful for target detection and more accurate results than the other one. Minimum noise fraction (MNF) transformation and pixel purity index (PPI) respectively were used to extract effectively endmembers. This research aims to identify the endmembers of the alteration minerals according to the ASTER image. Therefore, initially this study calculated image noise estimates using the MNF algorithm (Green et al., 1988). PPI was carried out to extract the endmembers from the data cluster, performed in the MNF space. The PPI algorithm detects the most spectrally pure or extreme pixels in hyperspectral images (Boardman, 1993; Boardman et al., 1995; Boardman and Huntington, 1997;  Kruse, 1997).In this research, the best result for PPI was obtained with 10000 iteration. The spectra of minerals from the USGS spectral library resampled to the ASTER bands were visually compared with the spectra of the extracted endmembers. Then, the four endmembers were identified by the present work as to the most similarity was observed between them and the spectral properties of alunite, kaolinite, muscovite and chlorite minerals (Fig. 2), one spectrum for each mineral.

Fig. 2: The reflectance spectra of extracted endmembers from ASTER image and USGS spectral library. (a) Alunite; (b) Kaolinite; (C) Muscovite; (D) chlorite. (Red line: USGS spectral library; Black line: ASTER scene).

5. Matched filtering algorithm

Matched filtering (MF) is a process that maximizes the response of a known endmember that is embedded in noise. MF is designed to derivate the maximum signal to noise ratio (SNR) of a signal that has been contaminated by noise (Harsanyi and Chang, 1994). The mathematical definition of the MF is expressed in Equation 1:

MFx=t-mTs-1(x-m)                                                                                                   (1)

Where

xis the sample vector, t the target vector,

mthe background mean and

sthe background covariance (Schott, 2007). MF score is computed for each pixel, the range of MF score is from 0 to 1 Score near one represent that the Spectra closely matched with the training spectrum while MF values of zero and lower represent background (no target) (Mundt et al., 2007). To specify the pixels that contain the endmember, a threshold value should be used to the MF result of every endmember. Firstly, this work used MF algorithm as spectral mapping method. The images were processed by The ENVI (RSI, 2000) software package. To obtain the optimal threshold of the kaolinite, alunite, muscovite and chlorite their histogram was plotted according to the MF results. The best accuracy for the classifier based on field data occurs at a threshold of >0.41, 0.4, 0.35 and 0.4 were defined for the kaolinite, alunite, muscovite and chlorite, finally 1740, 1699, 2294 and 1783 pixels were identified as the kaolinite, alunite, muscovite and chlorite target respectively.

Fig. 3: Alteration minerals map distinguished by MF with defined threshold value, a) kaolinite (red), b) alunite (green), c) muscovite (blue), d) chlorite (yellow).

The distribution maps of the alteration minerals were produced applying the thresholds on MF score (Fig. 3). The virtual verification (King and Clark, 2000) was applied to appraise the results of mineral distribution mapping. King and Clark (2000) introduced two kinds of verification of remote sensing imagery information: virtual and in situ. Virtual verification could be performed by examining the remote sensing data directly. In this study, the virtual verification is first used for MF results and their spectra visually compared with the USGS spectral library. Visual analysis of image spectra was done to check the presence of the absorption and reflection bands characteristics of the library spectrum corresponding to the endmember. Further verification was also considered by observations of thin sections and X-ray diffraction results presented by Mojedifar et al. (2013) and Honarmand et al. (2011). According to the Fig .3, there is no guarantee that the pixels identified by the thresholding process are correct. For an example in Fig .4, when taking 0.40 as the threshold for chlorite, the pixel with MF score of 0.5043 is considered as the target but the existence of the endmember in the pixel is refused by the virtual verification, this is falsely-mapped pixel, as well as the pixel with MF score of 0.3482 considered as background but the virtual verification confirms the existence of the endmember in that pixel, this is not-mapped pixel. Therefore, the major problem with thresholding process is that it considers only the intensity, not any relationships between the pixels and the reference spectra. It means that the thresholding process represents a contiguous range of MF value as answer and unable to introduce singular MF value or non-contiguous range. Then, it is expected that the results face such error when consider the thresholding process. Recent researches attempted to consider the issue. Molan et al. (2014) introduced the moving threshold technique to find the suitable value for optimal threshold. But the moving threshold is unable to choose discrete value from whole input data; it means that it is incapable to select the true discrete pixels from value above the threshold or below the threshold and just select the group of pixels that contain a value more than the determined threshold. Another way to look at the problem is that taking threshold don’t represent a good resolution for each alteration zone, in other words the obtained alteration zones were not discriminated in Fig .3 and what makes thresholding difficult is that these results usually overlap. This is the main problem encountered in image classification algorithms to map the distribution of alteration due to the similar spectral properties of the minerals in both altered and unaltered rocks. Therefore, this research intends to develop a different procedure to separate the alteration zones based on MF algorithm and reduce the errors in classification.

Fig. 4: Misclassification in MF result of chlorite with threshold value of 0.4.

6. Directed Matched Filtering (DMF) algorithm

As shown in the previous section, MF using thresholding process could not reliably discriminate the alteration zones. To address the issue, this paper introduces the directed matched filtering (DMF) algorithm in which a spectral signature-based filter is used instead of the thresholding process. DMF tries to find appropriate target pixels on the ASTER images with MF score > 0. It is obvious that the pixels with MF value <0, do not contain endmember, should not be considered by the DMF. The spectral signature-based filter is a spectral protocol that is defined by the expert to describe spectral signature of specific minerals of each alteration. This filter was developed for chlorite, kaolinite, alunite, and muscovite minerals to evaluate the resolution of DMF for alteration zones. The first step in using DMF is to make the detailed rules for the implementation of the spectral signature-based filter of DMF. The filter rules are defined as logical regulations to describe the spectral signature of the endmembers and should be designed by expert to control the outputs. The defined features and the spectral signature of each alteration mineral are illustrated in Fig .5.

Fig. 5: The spectral signature plot and the defined features (rules) alteration minerals (endmembers).

Absorption and reflection bands’ position, shape, number, and strength are considered to provide spectral signature-based filter of DMF for each mineral. The features showed in Fig. 5 indicate a numeral protocol about identification of image endmember spectra. Since a spectra in general is of mixtures of a variety of materials, the diagnostic features of more than one mineral could be observed in a spectrum. At the same time, a mineral with less evident spectral signature may be always invisible in all endmembers. Therefore this work attempted to impose the strict rules on spectral signature plot in Fig .5 although it may even provide an alteration region in discrete pixels. To consider these concerns with ASTER image data, restrictions, such as determination the upper and lower limits of the tolerance, were applied to diagnostic spectral signatures as displayed in Fig .5. The overall methodology which was adopted as DMF for this work was illustrated in the Fig .6.

Fig. 6: The flow diagram of DMF methodology.

Based on DMF, the MNF transformation is performed on 9 bands of ASTER image to reduce the dimensionality of the data. Then, they are applied to determine the pure pixels in the ASTER image using PPI technique. The spectra of pure pixels are plotted into n-dimensional scatter plot to define the endmembers. The spectra of minerals from the USGS spectral library is resampled to the ASTER bands and it is applied by spectral analyst to determine the material of the endmembers extracted from N-D Visualizer. MF algorithm is performed and all pixels of a special mineral with MF score > 0 are measured according to the corresponding protocol and the pixels with the highest matching are selected as target pixels. DMF algorithm was developed in a MATLAB code and the pseudocode of DMF is presented below:

DMF algorithm with the highest matching (Pseudocode).

Inputs: The pixels with MF value >0 extracted from MF result for each mineral (MF score); corresponding ASTER image pixels in 9 band for each mineral (Band); MF result for each mineral (MF).

Output: Directed Matched Filtering grayscale image.

START

Load MF score & Band

[minimum, absorption band]=min (Band)     % find minimum value and identify absorption band

                                                                             number in Aster image.

[maximum, reflection band]=max (Band)     % find maximum value and identify reflection band

                                                                             number in Aster image.

FOR(

i=1;i=length(Band‏‏) )                % starting with first pixel, iterate for every pixel in   

                                                                           ASTER image.

IF absorption band (: , i) == maximum absorption band number & reflection band (: i) ==

maximum reflection band number  % find the pixel with same absorption & reflection

                 band based on endmember spectral with the highest matching as defined in this paper.

IF Band (i, each band(between 1 to 9)) == The defiend rules in Fig .5.

% find the pixels with features similar to endmember and the highest 

                                 matching as defined in this paper.

target band(i , :)=Band(i ,:)         % put all band with the defined rules in the target band matrix.

target Mf value(i,:)=MF score(i,:)  % put all MF value with the defined rules in target MF

                                                           value matrix.

END

END

END

target MF=nonzeros (target Mf value) % remove the zero values from target MF value matrix.

MF image= imread (MF.tif) % read MF result obtained by ENVI software and put in MF image.

[j , k]=size (MF image)

[m , n]= size (target MF)

FOR (

r=1;r=j) % starting loop in the MF result in terms of the row.

FOR  (

c=1;c=k) % starting loop in the MF result in terms of the column.

      FOR  (

d=1;d=m) % starting loop in the target MF result.

                              IF MF image (r , c) == MF(d ,1) % find the correct endmember pixels in the

                                                                               MF result image.

% producing the target pixels of DMF output.

BREAK

ELSE

% producing the remian pixels of DMF output.

END

END

END

END         % producing the final output of DMF.

7. Results and discussion

1. Chlorite mapping

According to the MF results, the number of pixels with MF value >0 is 289,619 for chlorite. As discussed above, the visual interpretation of MF results (Fig. 3) Showed some misclassifications in chlorite mapping. To avoid this misassortment, DMF algorithm was developed by the present work. In DMF algorithm, if the actual reflectance spectra of pixels with MF value >0 satisfy the defined regulations of the spectral signature-based filter then it is considered as target pixel. Based on the defined filter of chlorite mineral, only 11 pixels were matched with the rules from 289619 pixels as input. The obtained pixels were listed in Table 1.

Table 1: The identified pixels as chlorite minerals by DMF algorithm.

The result map of DMF algorithm for chlorite is illustrated in Fig .7b (yellow points). DMF displays the pixels with the highest possible level of accuracy. On the other side, the value of every band is checked with the defined band feature of chlorite filter and the best matched pixels are considered as target pixel. Therefore, it is clear that the spectral signature of the identified pixels mapped in yellow complies with the rules outlined in the DMF filter of chlorite and shows highest possible level of matching in accordance to the endmember spectral signature.

Fig. 7: DMF results for the chlorite mineral (yellow pixels); a) filter with lowest possible level of matching; b) filter with highest possible level of matching.

In order to investigate sensitivity of the DMF results, the present work developed the spectral signature-based filter by changing the rules so that the lowest possible level of matching is observed in the detected pixels. The developed filter uses only the absorption and reflection bands which represents the spectral fingerprint of endmembers. Fig. 8 is illustrated the conditions of new filter against the former one for chlorite. As could be seen in Fig. 8a, there are only two restrictions that assess the actual spectrum of pixels. The maximum reflection band (5) and maximum absorption band (8) are two main rules to be applied by DMF for evaluation of chlorite pixels with MF value >0. For chlorite mineral, DMF algorithm with new filter was accomplished and 119 pixels were recognized. The identified pixels of chlorite are displayed in Fig. 7a with yellow points.

Fig. 8: The different levels of matching for chlorite mineral, a) filter with lowest possible level of matching; b) filter with highest possible level of matching.

 

Although the DMF result is produced by the lowest matching term, acceptable matching of identified pixels is confirmed by the virtual verification. The difference is that only 11 pixels from 119 pixels identified by the lowest filter show the higher purity. The high purity refers to a high amount of chlorite mineral in a single pixel. If the frequency distribution of MF value of pixels detected by the lowest filter is represented in a histogram (Fig. 12a), it could be told that the thresholding process in MF algorithm loses almost half of the pure pixels.

Fig. 12a displays simply the conspicuous distinction between MF and DMF algorithms. As shown in Fig .12a, the threshold value of MF was adjusted at 0.40 to determine the pixels that contain the endmember. This value classifies the gray scale image of ASTER into background and target pixels according to its value falls below or above the threshold, respectively. But the above histogram is for a distribution that is skewed to the left side of threshold boundary. This means the dominant frequency of correct pixels is observed in less than 0.40 and also based on Fig. 4 the presence of the endmember in the pixels was confirmed by the virtual verification. Therefore, selecting a contiguous range of pixels by MF does not seem to be effective because a large number of pure pixels were lost.

2. Other minerals mapping

Summary of the target pixels frequency obtained from different analysis performed in this study is represented in Table 2. Also, the results of developed algorithms for kaolinite, alunite and muscovite minerals are shown in Fig. 9, Fig. 10 and Fig. 11. According to the results, the histograms (Fig .12) show the highest frequency of target pixels in less than threshold boundary of MF algorithm in all minerals. The reason is that the pixels are just judged by its MF value in the final step of MF algorithm without any other consideration. The main risk associated with MF algorithm is concerned with the thresholding process and the contiguous range of which it is managed. The contiguous range is not an appropriate solution but is essentially subject to risks both unmapped pixels before the threshold boundary and falsely mapped pixels after it. The risk involved in the thresholding process could be found in the illustrated histograms. Especially about alunite mineral it is obvious that taking a special threshold value (0.40) by MF leads to loss of the high frequency target pixels.

 

Table 2: The summary table of identified pixels by MF and DMF algorithms.

Kaolinite and muscovite minerals have the same number of pixels as those on the Table 2 by DMF with lowest filter. Its cause could be ascribed to structure of lowest filter, the defined rules by which it is formed. The spectral attributes of the selected minerals were visually investigated using the USGS spectral library. This visual interpretation of the absorption bands’ number, strength, position, and shape led to determine the spectral signature-based filter for each selected mineral. Since kaolinite and muscovite have the same spectral characteristics, there is no possible to discriminate them by the lowest filter. Therefore, DMF with lowest filter is especially useful where dissimilar minerals are involved. But this issue has been addressed by DMF with highest filter and it could differentiate between kaolinite and muscovite minerals according to its spectral regulations.

Fig. 9: DMF results for the alunit mineral (green pixels); a) filter with lowest possible level of matching; b) filter with highest possible level of matching.

Fig. 10: DMF results for the muscovite mineral (blue pixels); a) filter with lowest possible level of matching; b) filter with highest possible level of matching.

Fig. 11: DMF results for the kaolinite mineral (red pixels); a) filter with lowest possible level of matching; b) filter with highest possible level of matching.

Fig. 12: The MF value distribution of alteration minerals pixels with the lowest possible level of matching; (a) chlorite, (b) alunit, (c) muscovite, (d) kaolinite.

8. Validation of the results

This study sought to discriminate the distribution of three alteration zones over the seven copper–mineralization districts by using two kinds of DMF algorithm. A detailed picture of the spatial distribution of the studied alteration minerals in the northwestern part of the KCMA area is provided by the DMF algorithm based on the lowest and highest filter (Fig .13). In these Figures each alteration zone was allocated a unique colour and was exhibited over a gray scale image. Muscovite mineral is depicted in blue, kaolinite in red, alunite in green and chlorite in yellow. The virtual verification confirmed the presence of the endmember in the pixels identified by DMF algorithm. Fig. 14 shows that the reflectance spectra which were extracted from identified pixels in highest filter of DMF are overlapping the spectra from the USGS spectral library. Also to verify the lowest filter, 35 pixels were randomly selected and made an illustration of the spectral reflectances for alteration minerals. As outlined in Fig. 15, the overall structure of spectral signature for each mineral is approved by virtual verification (i.e. maximum reflection and minimum absorption). But the lowest filter curves are smoother than the highest filter ones because of different levels of matching by defined filters. It means the pixels that contain more than one cover type are more found in the lowest filter outputs.

Fig. 13: Distribution map of the alteration minerals with (a) highest filter and (b) lowest filter of DMF; kaolinite (red), muscovite (blue), alunite (green) and chlorite (yellow).

Fig. 14: Virtual verification of the pixels identified by highest filter of DMF; (a) alunite, (b) chlorite, (c) kaolinite and (d) muscovite.

 

 

Fig. 15: Virtual verification of the pixels identified by lowest filter of DMF; (a) alunite, (b) chlorite, (c) kaolinite and (d) muscovite.

Mojedifar et al. (2013) and Honarmand et al. (2011) checked the altered areas in the field and in the laboratory by observations of thin sections and X-ray diffraction analysis. Their studies showed that sericite alteration is dominant at the Iju, Serenu, Chahfiroozeh, Meiduk, Parkam, Kader, and Abdar porphyry copper deposits. Two kinds of phyllic alteration could be recognized in the field including ferric-iron-rich and iron-oxide poor phyllic alterations. The iron-oxide-rich phyllic zone shows a large amount of iron oxide minerals on the surface. The common secondary minerals at the Kader, Iju, Serenu, Parkam, Meiduk, and Abdar deposits are in the form of goethite, jarosite and minor hematite. Three zones of hydrothermal alteration are comparatively uniform over the Kader area that consists of phyllic, argillic, and propylitic alteration. Because of the difficulties in discrimination of clay minerals in thin section, they analyzed the rock samples by spectroradiometer. Argillic alteration is present in the deposits at Kader, Serenu, Meiduk, Parkam, Godekolvary and Abdar. Propylitic alteration happens around most of the mineralized areas. Fig .13a shows the result of DMF with the highest filter. A comparison of the altered areas in the image with field data reveals strict classification in the identification of alteration minerals. It means that the observed alteration zones are less than expected. As seen, from seven alteration areas in the region, the Kader deposit is acceptably identified compared with the actual extent on the ground. In this region, three types of alteration pixels are observed that consists of phyllic, argillic and propylitic. Because of exacting rules of highest filter in DMF, only the Kader deposit was highlighted by the pixels of muscovite(blue) and kaolinite(red) and the other deposits could not be differentiated in Fig. 13a while the whole known deposits could be found in lowest filter state (Fig.  13b). The broad phyllic and argillic zones in the area are characterized by muscovite (sericite) and kaolinite, respectively. Field studies reveal that the distribution of areas of phyllic/argillic alterations classified by DMF are related to seven known deposits. Although the same spectral signature of kaolinite and muscovite leads to overlapping with each other as observed in Fig .13b using lowest filter. The advanced argillic zone could be indexed by alunite which shows an Al-OH absorption feature near 2.17 µm (ASTER band 5, Fig .5); they show considerable different spectral shapes to those of muscovite. Kaolinite shows a secondary feature or shoulder at 2.20 µm (ASTER band 6, Fig .5). Based on Fig .13b , advanced argillic pixels are more observed at the Abdar and Godecolvari deposits. The propylitic alteration, which is characterized by chlorite, could be found out more around the mineralized areas as seen in Fig .13b.

9. Conclusion

A comprehensive alteration minerals map was acquired by the analysis of the ASTER images in the KCMA area. The MF method was used to analyze the ASTER data to map the image-derived endmembers. This study challenged the results of MF method using thresholding process and introduced a new procedure (DMF) to classify the gray scale MF images to background and pure pixels. Then, the results of the DMF mapping were verified by the visual interpretation of the signature features and field studies. The comparison proved that, at first: selecting a contiguous range of MF values could not identify a desireable result, second: unexpectedly the considerable frequency of pure pixels could be observed in the MF values less than threshold value. However, the validity of the DMF results depends strongly on the defined filter and the expert could be adjusted the preferred level of accuracy. As stated earlier in the text, the high overlapping may be the result of some difficulties belongs to mapping of spectrally similar minerals using the lowest filter of DMF. However, this issue could be considered using the highest filter of DMF.

References

Abrams, M.J., Brown, D., Lepley, L., Sadowski, R., 1983. Remote sensing for porphyry copper deposits in southern Arizona. Economic Geology 78, 591–604.

Abrams, M., 2000. The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER): data products for the high spatial resolution imager on NASA’s Terra platform. International Journal of Remote Sensing 21, 847–859.

Abrams, M., Hook, S., Ramachandran, B., 2004. ASTER User Handbook, Version 2. Jet Propulsion Laboratory, California Institute of TechnologyOnline: https://asterweb.jpl.nasa.gov/content/03_data/04_Documents/aster_user_guide_v2.pdf

Azizi, H., Tarverdia, M.A., Akbarpourb, A., 2010. Extraction of hydrothermal alteration from ASTER SWIR data from east Zanjan, northern Iran. Advances in Space Research, 46, 99-109.

Bennett, S.A., Atkinson, W.W., Kruse, F.A., 1993. Use of thematic mapper imagery to identify mineralization in the Santa Teresa District, Sonora, Mexico. International Geology Review. 35, 1009–1029.

Boardman, J.W., 1993. Automated spectral unmixing of AVIRIS data using convex geometry concepts. Summaries of the 4th Annual JPL Airborne Geoscience Workshop, JPL Publication 93-26, vol. 1. Jet Propulsion Laboratory, Pasadena, CA, 11–14.

Boardman, J.W., Kruse, F.A., Green, R.O., 1995. Mapping target signatures via partial unmixing of AVIRIS data. Summaries, Fifth JPL Airborne Earth Science Workshop, JPL Publication 95-1, vol. 1, 23–26.

Boardman, J.W., Huntington, J.F., 1997. Mineralogic and geochemical mapping at Virginia City, Nevada, using 1995 AVIRIS data. Proceedings of the 12th International Conference on Applied Geologic Remote Sensing, vol. 1, Denver, Colorado, 17–19 November 1997. ERIM, Ann Arbor, MI, 191–198.

Bodruddoza, M.D., Fujimitsu, Y., 2012. Mapping hydrothermally altered mineral deposits using Landsat 7 ETM+ image in and around Kuju volcano, Kyushu, Japan. Journal of Earth System Science 121, 1049–1057.

Carranza, E.J.M., Hale, M., 2002. Mineral imaging with LANDSAT thematic mapper data for hydrothermal alteration mapping in heavily vegetated terrane. International journal of remote sensing 23, 4827-4852.

Cox, D. P., Singer, D.A., E.B., 1986. Mineral deposits models, U.S. geological survey bulletin, 1693.

Dimitrijevic, M.D., 1973. Geology of the Kerman region. Geological Survey of Iran publication, Tehran, Rep., 52, 334.

Di Tommaso, I., Rubinstein, N., 2007. Hydrothermal alteration mapping using ASTER data in the Infiernillo porphyry deposit, Argentina. Ore Geology Reviews, 32, 275-290.

Evans, A. M., E.B., 1993. Ore geology and industrial minerals: An Introduction, third ed. Blackwell Scientific Publications, 390.

Gabr, S., Ghulam, A., Kusky, T., 2010. Detecting areas of high-potential gold mineralization using ASTER data. Ore Geology Reviews, 38, 59-69.

Goosens, M.A., Kroonenberg, S.B., 1994. Spectral discrimination of contact metamorphic zones and its potential for mineral exploration, province of Salamanca, Spain. Remote Sensing of the Environment 47, 331–344.

Green, A.A., Berman, M., Switzer, P., Craig, M.D., 1988. A transformation for ordering multispectral data in terms of image quality with implications for noise removal. IEEE Transactions on Geoscience and Remote Sensing 26 (1), 65–74.

Harsanyi, J.C., Chang, C.I., 1994. Hyperspectral image classification and dimensionality reduction: an orthogonal subspace projection approach. IEEE Transactions on Geoscience and Remote Sensing 32, 779–785.

Hassanzadeh, J., 1993. Metallogenic and tectono-magmatic events in the SE sector of the Cenozoic active continental margin of Iran (Shahr-e-Babak area, Kerman province). Unpublished Ph.D. Thesis, University of California, Los Angeles, 204.

Heusdens, R., Engelen, S., Buist, J., Noroozi, A., Sundaramoorthy, P., Verhoeven, C., Bentum, M., Gill, E., 2012. Match filtering approach for signal acquisition in radio-pulsar navigation. 63rd International Astronautical Congress, Naples, Italy.

Honarmand, M., Ranjbar, H., Shahabpour, J., 2011. Application of Spectral Analysis in Mapping Hydrothermal Alteration of the Northwestern Part of the Kerman Cenozoic Magmatic Arc, Iran.  Journal of Sciences 22, 221-238.

Hubbard, B.E., Crowley, J.K., 2005. Mineral mapping on the Chilean-Bolivian Altiplano using Co-orbital ALI, ASTER and Hyperion imagery: data dimensionality issues and solutions. Remote Sensing of Environment 99, 173-186.

Kaufmann, H., 1988. Mineral exploration along the Aqaba–Levant structure by use of TM data; concepts, processing, and results. International Journal of Remote Sensing 9, 1639–1658.

King, T.V.V., Clark, R.N., 2000. Verification of remotely sensed data. In: Kuehn, F., King, T., Hoerig, B., Pieters, D. (Eds.), Remote Sensing for Site Characterization. Springer, Berlin, 59–61.

Knepper, D.H., Simpson, S.L., 1992. Remote sensing in Geology and mineral resources of the Altiplano and Cordillera Occidental, Bolivia. U.S. Geological Survey. 47–55.

Kruse, F.A., 1997. Regional geological mapping along the Colorado Front Range from Fort Collins to Denver using the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Proceedings of the 12th International Conference on Applied Geologic Remote Sensing, vol. II, Denver, Colorado, 17–19 November. ERIM, Ann Arbor, MI, 91–98.

Kruse, F.A., Bordman, J.W., Huntington, J.F., 2003. Comparison of airborne hyperspectral data and EO-1 Hyperion for mineral mapping. IEEE Transactions on Geoscience and Remote Sensing, 41(6), 1388-1400.

Livo, E., Clark, R.N., Knepper, D., 1993. Spview Spectral Plot program for accessing the U.S. Geological Survey Digital Spectral Library Database with MS-DOS personal computers. U.S. Geological Survey, Open File Report 93-593, 26.

Lowell, J. D., Guilbert, J. M., 1970. Lateral and vertical alteration-mineralization zoning in porphyry ore deposits. Economic Geology 65, 373-408.

Mars, J.C., Rowan, J.C., 2006. Regional mapping of phyllic and argillic altered rocks in the Zagros magmatic arc, Iran, using Advanced Space-borne Thermal and Reflection Radiometer (ASTER) data and logical operator algorithms. Geosphere 2, 161-186.

Mojedifar, S., Ranjbar, H., Nezamabadi, H., 2013. Adaptive neuro-fuzzy inference system application for hydrothermal alteration mapping using ASTER data. Journal of Mining & Environment, Vol.4, No.2, 83-96

Molan, Y.E., Refahi, D., Tarashti, A.H., 2014. Mineral mapping in the Maherabad area, eastern Iran, using the HyMap remote sensing data. International Journal of Applied Earth Observation and Geoinformation, 27,117–127.

MUNDT, J.T., STREUTKER, D.R., GLENN, N.F., 2007.  Partial unmixing of hyperspectral imagery: theory and methods. Proceedings of the American Society of Photogrammetry and Remote Sensing, Tampa, Florida, 46–57.

Porwal, A.E., Carranza, J.M., Hale. M., 2004. A Hybrid Neuro-Fuzzy Model for Mineral Potential Mapping. Mathematical Geology 36, 803-826.

Pour, A.B., Hashim, M., 2011. Application of Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data in geological mapping. International Journal of the Physical Sciences Vol. 6(33), 7657 – 7668.

Pour, A.B., Hashim, M., Marghany, M., 2014. Exploration of gold mineralization in a tropical region using Earth Observing-1 (EO1) and JERS-1 SAR data: a case study from Bau gold field, Sarawak, Malaysia. Arabian Journal of Geosciences, 7, 2393-2406.

Pour, A.B., Hashim, M., 2015. Hydrothermal alteration mapping from Landsat-8 data, Sar Cheshmeh copper mining district, south-eastern Islamic Republic of Iran. Journal of Taibah University for Science Vol. 9(2), 155-166.

Rowan, L.C., Schmidt, R.G., Mars, J.C., 2006. Distribution of hydrothermally altered rocks in RekoDiq, Pakistan mineralized area based on spectral analysis of ASTER data. Remote Sensing of Environment 104, 74-87.

RSI, 2000. ENVI User’s Guide, the Environment for Visualizing Images, Version 3.2. Research Systems, 2995 Wilderness Place, Boulder, CO 80301, USA, pp. 930.

Schott, J.R., E.B., 2007. Remote Sensing: The Image Chain Approach, second ed. Oxford University Press, New York, pp. 688.

Soheili, M., 1981. Geological map of Anar, 1:250 000 map. Geological Survey of Iran publication, Tehran.

Tangestani, M.H., Mazhari, N., Agar, B., Moore, F., 2008. Evaluating Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data for alteration zone enhancement in a semi arid area, northern Shahr-e-Babak, SE Iran. International Journal of Remote Sensing 29, 2833-2850.

Zarasvandi, A., Liaghat, S., Zentilli, M., 2005. Geology of the Darreh–Zerreshk and Ali-Abad porphyry copper deposits, Central Iran. International Geology Review, 47(6), 620–646.