WEST OF IRAQ SATELLITE IMAGE CLASSIFICATION USING FUZZY LOGIC

In this paper, land use/cover classification using fuzzy techniques which involves several steps, from designing the parameters of the membership functions through classification of the satellite image to the refining the final product. To decide the threshold parameters of membership functions that lead to appropriate classification of the scene, one band of landsat-5 were investigated by the features of the histogram of each area to be classified. The results of fuzzy system (Mamdani type) has been compared with the classical method (* Maximum likelihood classification *) and encourage us to use this technique for other bands with optimum rules for future works.


INTRODUCTION
Classification is the fundamental image processing task to extract information from remote sensing data.Both crisp and soft classifications may be performed.In a crisp classification, each image pixel is assumed pure and is classified to one class.Often, particularly in coarse spatial resolution images, the pixels may be mixed containing two or more classes.Soft classifications that assign multiple class memberships to a pixel may be appropriate for images dominated by mixed pixels.Both supervised and unsupervised approaches may be adopted [1].
Knowledge of both land used and land cover is important for economy planning of a region.While the land used related to human activities residential, institutional, commercial and recreational …etc., the land cover relate to the various type of features present on the surface of the earth.For      In determination whether the training areas that have been selected are well represented, histogram was used: if the histogram has a single peak, then the training area is distinct and there is no confusion between it and another training area unless they have the same gray level.
A histogram with a wide distribution would indicate that there may be an ambiguity between the current and some other region.
Since the signature separability showed that tree and vegeTable are very poorly separated (low values of Transformed Divergence; big overlap between the signatures of two classes).
Those some uncorrected result in the classification operation will appear.The signature statistics gave a list of each of the classes, with the mean values and standard deviations for the class selected.These data were used later in the definition of the membership function.As it can be seen in Figure (3), similar values (overlap) can be found in the used image for crop, tree and urban area classes.This is due to the similar characteristics in the spectral response (reflectance) of these classes in the wavelength range 0.5-0.59µm.

METHODOLOGY USED
Fortunately, they can be better separated cause of the bigger difference in other bands for future works.

CLASSIFICATION PROCEDURE
Since the goal of both procedures, maximum likelihood (ML) and fuzzy logic, is to classify the image, input data must be the same.That is, one band channel is used as the starting point for the image classification based on fuzzy logic.
In this paper, Matlab's Fuzzy Logic Toolbox is used which need two parameters for the valid membership function definition: mean and standard deviation values.
The Fuzzy Inference System (FIS) Editor displays general information about a fuzzy inference system: a simple diagram with the names of the input variable (B1 channel) and those of each output variable (Shallow water, Deep water, urban area, crop, tree and vegetation).There is also a diagram with the name of the used type of inference system.
The Membership Function Editor is used to display and edit all membership functions associated with the input and output variables for the entire fuzzy inference system.Because of the smoothness and non-zero values, in order to define a membership function, in the process of image classification simple Triangular function is used.Based on the descriptions of the input and output variables (Deep Water, Shallow Water, Urban, VegeTable, Crop, Tree, Bear and Unknown), the rule statements can be constructed in the Rule Editor.
When the variables have been named and the membership functions have appropriate shapes and names, everything is ready for writing down the rules.
Rules for image classification procedure in verbose format are as follows: At this point, the fuzzy inference system has been completely defined, in that the variables, membership functions and the rules necessary to calculate classes are in place.Classification is conducted by the Matlab's m-file.

Output
images coming from maximum likelihood classification (using TNTmip2010software) shown in figure 6 and fuzzy classification (using Matlab) shown in Fig. 7 and Fig. 8 can be compared.
These gray scale images are produced in such way that pixels coming from the same class have the same digital numbers in both mages: Deep Water(1), Shallow Water(2), Tree(3), VegeTable(4), Urban(5), Crop(6), Bare (7) and Unknown(8).This is the basis for image comparison.Percentage of classified pixels in both methods is given in the Table 4. (Overall number of image pixels is 262144)..6 related to Fig. 9 show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them..7 related to Fig. 10 show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them.
From comparing the similarity from the previous results it can be show that membership function of type Gaussian with mean and standard deviation values which used with Mamdani type of Fuzzy Inference System is the best one for classification purpose in our study case with 73.3967% similarity.While the worst result was membership function of type Triangular with peak and standard deviation with 44.3829% similarity.

CONCLUSIONS AND FUTURE WORKS
In conclusion there are several points: proper planning exercise information on both the above aspects should available separately.The satellite based remote sensing has been very popular and different countries have lunched their remote sensing satellite for this purpose.The collected data are processed and interpreted in different forms using digital techniques or optical techniques.Although the visual interpretation of image is being used in many applications, it does not interpret the image pixel by pixel, instead it provide aggregated information related to image features of known objects.As a consequence, the information results for land used and covered provided by human interpreter is less accurate and overlapping in many places[2].
most common source of satellite-based remote sensing data available to the civil engineer.Landsat-5 thematic mapper (TM) launched on March 1, 1984, The TM is a remote sensor for acquisition of data in seven bands, and the wavelength range and location of the TM bands have been chosen to improve the spectral different abilities of major Earth surface features.
Fig. (1) Satellite image show the location of the study site.

Figure
Figure 2. Training areas

Figure
Figure 3. Classes overlapping in their Histograms

Figure
Figure 10.a Classified image (Gaussian with Peak and Std Values)

Fig. 7 .
Fig. 7.a show the result of fuzzy classification using the Triangular membership function with mean, min and max values.Fig 7.b show the similarity (White point) between fuzzy logic classification and ML classification with 71.5206%.While Table.4 related to Fig.7show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them.

Fig. 8 .
Fig. 8.a show the result of fuzzy classification using the Triangular membership function with peak and standard deviation values.Fig 8.b show the similarity (White point) between fuzzy logic classification and ML classification with 44.3829 %.While Table.5 related to Fig.8show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them.

Fig. 9 .
Fig. 9.a show the result of fuzzy classification using the Gaussian membership function with mean and standard deviation values.Fig 9.b show the similarity (White point) between fuzzy logic classification and ML classification with 73.3967%.While Table.6 related to Fig.9show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them.

Fig. 10 .
Fig. 10.a show the result of fuzzy classification using the Gaussian membership function with peak and standard deviation values.Fig 10.b show the similarity (White point) between fuzzy logic classification and ML classification with 66.0923%.While Table.7 related to Fig.10show the percentage pixels of each class in both fuzzy and ML and the absolute difference between them.

Table 7 Percentage of classified pixels using Gaussian with Peak and Std values
1. Gaussian membership function with meanand standard deviation values has best smooth and training area separability which cover the most universe of discourse yields to good results in our work.the band histogram range in our study area have a small gap between training area gray level value which effect the fuzzy accuracy classification in one way or Since