A gray level image with equal amount of red, green and blue per pixel. A false color conversion is done by changing colors in the RGB display [15] [16] . False color combination (FFC) is often performed to emphasize the effects or special effects. It’s better to use Non-visible bands in false color combination (FFC) and they have more values than single-band images [16] . Each composite band is made of three colors and each color is assigned to a band. Before obtaining composite band using OIF index, 84 triplex composite bands modes should be considered for the 9-band ASTER sensor. In fact OIF index used to select a combination of three bands which is calculated by dividing the three-bands standard deviation to the correlation coefficient absolute value (ABS (CCi)) [17] [18] . The Image Optimum Index Factor is written in Equation (1). The Image Optimum Index Factor is written in Equation (1). In Table 2 the OIF composite band is shown and we can see it in Figure 3. In this figure green Gabbros, red-black dunite and dark green bluish Chromite are obvious.

*The highest OIF is 0.37210387 for RGB: (B3, B4, B8).

Spectral rationing is a multi-spectral image processing method which includes dividing one band to another one, (usually after initial corrections for atmospheric path radiation or sufficient movement imposed by the multi-spectral sensor) [19] [20] . Sometimes these images used in Mineralogy index calculation too. They apply in calculating the considered indexes too. Generally, in the band ratio method by recognizing absorption and reflection spectrum specifications of mineral, good bands with the most reflection and absorption are selected. In Figure 4 resampling Spectral diagram the Peredotit-2, Peredotit-4, Pyroxenite,

and Serpentine are shown. In this paper, we use the composite band of band ratio (R: (B2 + B4)/B3, G: (B5 + B7)/B6, B: (B7 + B9)/B8) proposed by Amer in 2010 and from now on we call it Amer band ratio [21] . It is shown in Figure 5 where dark pink Chromite, yellow pink Gabbro, orange and pale green Carbonates and dark green vegetation are displayed. In parts B, C and D we can see the threshold for each part of the Amer band ratio. Threshold function used in many types of Spectral minerals algorithms such as procedures based on the interpretation of Spectral knowledge, methods to identify the ultimate goal or the distribution of pixels extractive for maping minerals and rocks. The threshold calculation statistical methods lead to low concentrations removal [22] . At first, the pixel value of each image sets from zero to 255 and then threshold calculation is done using Equations (2)-(4) with 98%, 95% and 92% confidence level. In Figure 5 the threshold for Amer’s band ratio is specified.

ASTER sensor’s spectral library includes data from three spectral library like Johns Hopkins University (JHU), NASA Jet Propulsion Laboratory’s spectral library (JPL) and United States geological Survey Headquarters (USGS). For more information you can refer to the website (https://speclib.jpl.nasa.gov/search-1) which contains the whole Spectrum, Stone, Soil, Mineral, Vegetation and a comprehensive collection of more than 2300 covering products with a wavelength range of 0.4 - 15.4 micrometers [23] . In fact, we choose Peredotit-2, Peredotit-4 rock spectrum (two Peridotite stone with different combinations marked by the numbers 2 and 4), Pyroxenite, and Serpentine from spectral li-

brary; after that we should specify reflection, absorbtion band according to the spectrum. Lisweanit’s reflection and the absorption band are taken from Sankaran Rajendran article [24] . Principal Component’s conversion is one of the most popular tools to reduce the dimensionality [25] .

Generally, it is a linear transformation maximizing the variance by transferring data to a new coordinate system. In Feature Oriented Principal Components Selection (FPCS) we make use of good and well-known bands with appropriate information. In this way, only basic information about minerals, stone and vegetation properties are required and it’s based on principal component conversion capabilities in mapping and data variance details in consecutive components. This method is known as Crosta [26] , so that 4-bands are selected as 2 absorption and 2 reflection bands (Based on the spectral properties of minerals and Stone). In this method, according to evaluating eigenvector, we can determine which of the principal components reflect the information about spectral signature of specific purpose [27] . The choice of PCs is appropriate in that, in Vector matrix value, the bands absorption and reflection of the target minerals have the highest values at the same time, while these values are also non-symbols [28] [29] . Table 3 shows the Lisweanit stone’s eigenvalues matrix and you can see the eigenvalues matrix for Peredotit-2, Peredotit-4, Pyroxenite and Serpentine. Table 4 contains threshold values for the FPCS of Peredotit-2, Peredotit-4, Pyroxenite, Serpentine and Lisweanit. The selected principal component analysis (FPCS) images are shown in Figure 6.

*EV: Eigenvector.

*Thershold Value, **Absorption, ***Reflection, Conf: Confidence, L: Low, H: High.

After identifying stones using FPCS method, determining the images accuracy is absolutely essential because indicates that how well the resulting map corresponds with the ground truth. In this project Davoudzadeh map is used to verify the validation [1] . Sampling is done at the first phase and then we perform validation.

・ Sampling: Generally, in this study, random sampling pattern was used. In addition to random mode maintenance, this pattern solves the nonuniform distribution of points [30] . In fact, at first, we identify Statistical Society pixels on the basis of the impact area and spatial resolution. Justice and Townshend in 1981 proposed a useful formula for determining each pixel sample locations area [30] [31] . In Equation (5), A is minimum sample site dimension, P is image pixel dimension and L refers to estimated locational accuracy in number of pixels which is accurately locate an area on the ground to ±0.5 pixel.

The amount of A is divided by effect area to determine the population. The maximum number of samples is obtained according to sample size. The Equation (6), Cochran’s rule, is one of the most widely used methods for sample size calculation.

Using Cochran formula is the easiest way determining the sample size [32] . The formula elements are as follows: N: Size of population (the population of the city, province, etc), n: Number of samples, P: proportion of the population lacks certain traits (e.g. male population), t or z: Percent confidence level of acceptable standard error, d: degree of certainty or potential efficiency, 1 − p = q proportion of the population lacks certain traits (e.g., women), p and q usually consider as 0.5, Z value is usually 1.96, d may be 0.01 or 0.05.

・ Validation: No method is invoked yet its accuracy has been tested. Therefore, accuracy assessment is done to ensure the variety of performed procedures [33] . Validation indicates the extracted maps confidence level and it must be at least 85% in the images derived from spectral method [34] . Verification means assessing the conformity of images derived from spectral method with ground truth [35] [36] . In this study, we use KAPA and overall accuracy methods.

・ Error matrix and Kappa coefficient: Error matrix is an efficient tool for the preparation and presentation of accuracy assessment information. Sometimes, it’s called contingency matrix or probability table. Error matrix compares the known relation between reference data (ground truth) and the relevant results of the automated classification [37] . It is a square matrix where the number of rows and columns equal to the number of categories which their classification accuracy was assessed.

Beside, Kappa coefficient or statistical kappa coefficient is an accuracy evaluation criteria. KAPA index or K is the result of Kapa analysis shown as an index to measure the classification accuracy based on the difference between error matrix accuracy and accuracy changes by sum of rows and columns [38] .

In fact, Equation (7) K index indices the disorder between real consistency in reference data and an automated classifier or possible consistency between reference data and random classifier.

When the actual agreement is close to one and expected agreement is close to zero, K value comes to one, so it is the ideal mode [34] . In fact, the Kappa index value is between zero and one, if the amount of kappa is zero indicates a completely random classification and if the amount of kappa is one indicates a quite accurate classification 100% in accordance with the ground truth [35] [36] . Overall accuracy is the ratio of properly sorted out pixels to total known pixels. Equation (8) indicates the overall accuracy. In this regard is sum of diagonal elements and N indices the total pixels or set of elements [38] . In Table 5 the number of points has been sampled using Davoudzadeh geological map based on the Cochran formula and Table 6 shows the validation results.