GISdevelopment.net --> Application --> Miscellaneous
Haiza Wahida Haron & M. Rafee Majid
Dept. of Urban and Regional Planning
Universiti Teknologi Malaysia
81310 Skudai, Johor, Malaysia
Emails:haiza_net@hotmail.com, rafee@utm.my
Abstract:
This paper compares three main methods of classification in satellite imageries (Quicbird). These classifications are based on fuzzy theory (Mamdani method), neural network and minimum distance. These methods are supervised and implemented in Mat Lab. In this paper an image is divided into three classes including vegetation, urban regions and water land. Only spectral and radiometric characteristic of image pixels are considered and geometrical and topological relation between pixels aren’t used.
1. Introduction
Generally a satellite image contains different features. The goal of image classification is to divide an image into regions which have same characteristics. Manual classification is done with high precision but development of software technology and expert systems promote us to have more precision with high speed and finally less expense .Using a classified image in geographical information system (GIS) results to better interpretation and decision making. In this paper the three main classification approaches including neural network, fuzzy theory and minimum distance are compared.
The rest of the paper is organized as follows. Section 2 describes the concept of pattern classifiers based on fuzzy if–then rules. Section 3 introduces our neural network methods that use for classifying an image. Section 4 then details minimum distance that used Sections 5 shows results of these methods for comparing that which one is better for this image and finally sees the conclusion using of this methods.
2. Fuzzy classification
Fuzzy logic is relatively young theory. Major advantage of this theory is that it allows the natural description, in linguistic terms, of problems that should be solved rather than in terms of relationships between precise numerical values. This advantage, dealing with the complicated systems in simple way, is the main reason why fuzzy logic theory is widely applied in technique. It is also possible to classify the remotely sensed image (as well as any other digital imagery), in such a way that certain land cover classes are clearly represented in the resulting image. In this paper, a priori knowledge about spectral information for certain land cover classes is used in order to classify Quickbird image in fuzzy logic classification procedure.
input (image channels) and output variables (land classes) are introduced in Mat lab’s environment,
Membership functions are defined using results from supervised classification, Mat lab’s Fuzzy Logic Toolbox was used in definition of fuzzy logic inference rules, These rules are tested and verified through the simulation of classification procedure at random sample areas and at the end,
Quick bird image classification was conducted.
2.1 Fuzzy inference system
Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. The process of fuzzy inference involves: membership functions, fuzzy logic operators and if-then rules. There are two types of fuzzy inference systems that can be implemented in the Fuzzy Logic Toolbox:
Mamdani-type and
Sugeno-type.
Mamdani's fuzzy inference method is the most commonly seen fuzzy methodology and it expects the output membership functions to be fuzzy sets. After the aggregation process, there is a fuzzy set for each output variable that needs defuzzification. Sugeno-type systems can be used to model any inference system in which the output membership functions are either linear or constant. This fuzzy inference system was introduced in 1985 and also is called Takagi-Sugeno-Kang. Sugeno output membership functions (z, in the following equation) are either linear or constant. A typical rule in a Sugeno fuzzy model has the following form:
If Input 1 = x and Input 2 = y, then Output is z = ax + by + c
For a zero-order Sugeno model, the output level z is a constant (a=b =0). [2]
2.1.1 Membership function
Membership function is the mathematical function which defines the degree of an element's membership in a fuzzy set. The Fuzzy Logic Toolbox in Mat lab’s environment includes 11 built-in membership function types. These functions are built from several basic functions:
Piecewise linear functions,
The Gaussian distribution function,
The sigmoid curve and
Quadratic and cubic polynomial curve
2.1.2 Fuzzy logic operators
The most important thing to realize about fuzzy logical reasoning is the fact that it is a superset of standard Boolean logic. In other words, if the fuzzy values are kept at their extremes of 1 (completely true) and 0 (completely false), standard logical operations will hold. That is, A AND M operator is replaced with minimum - min (A,M) operator, A OR M with maximum - max (A, M) and NOT M with 1-M and others.
2.1.3 If-Then rules
Fuzzy sets and fuzzy operators are the subjects and verbs of fuzzy logic. Usually the knowledge involved in fuzzy reasoning is expressed as rules in the form: If x is A Then y is B where x and y are fuzzy variables and A and B are fuzzy values. The if-part of the rule "x is A" is called the antecedent or premise, while the then-part of the rule "y is B" is called the consequent or conclusion. Statements in the antecedent (or consequent) parts of the rules may well involve fuzzy logical connectives such as ‘AND’ and ‘OR’. In the if-then rule, the word "is" gets used in two entirely different ways depending on whether it appears in the antecedent or the consequent part.
2.2 Classification procedure
The Fuzzy Inference System (FIS) Editor displays general information about a fuzzy inference system: a simple diagram with the names of each input variable (green, red and blue channel) and those of each output variable (water, urban area, vegetation). The Membership Function Editor is used to display and edit all membership functions associated with all of the input and output variables for the entire fuzzy inference system. Figure (1) shows original Quick bird image satellite and figure (2) depicts classified image using fuzzy logic.
Figure1. Quikbird satellite image
Figure 2. classified image with fuzzy logic
Figure1. Quikbird satellite image Figure 2. classified image with fuzzy logic
3. Neural Network
Artificial neural networks have been employed for many years in many different application areas such as speech recognition and pattern recognition .In general, these models are composed of many nonlinear computational elements operating in parallel and arranged in patterns reminiscent of biological neural nets. Similar to pattern recognition, there exist two types of modes for neural networks – unsupervised and supervised. The unsupervised type of these networks, which possesses the self-organizing property, is called competitive learning networks. A competitive learning provides a way to discover the salient, general features which can be used to classify a set of patterns. Because of the variations of object characteristics, atmosphere condition, and noise, remotely sensed images may be regarded as samples of random processes. Thus, each pixel in the image can be regarded as a random variable. It is extremely difficult to achieve high classification accuracy for most per-pixel classification algorithms (classifiers). Photo interpreters have had pre-eminence in the use of context-dependent information for remote sensing mapping. Neural networks have been recognized as an important tool for constructing membership functions, operations on membership functions, fuzzy inference rules, and other context-dependent entities in fuzzy set theory. In this study, the competitive learning neural networks for the image classification.[10].
3.1 ARTIFICIAL NEURAL NETWORKS FOR MULTISPECTRAL IMAGE CLASSIFIERS
Artificial Neural Networks (ANNs), a brain-style computation model, have been used for many years in different application areas such as vector quantization, speech recognition and image classification. In general, ANN is capable of tolerating the noise, distortion and incompleteness of data taken from the practical applications. Researchers have developed several different paradigms of ANNs. These paradigms are capable of detecting various features represented in input signals. An ANN is usually composed of many nonlinear computational elements. These computational elements operate in parallel to simulate the function of human brain. Hence, an ANN is characterized by the topology, activation function, and learning rules. The topology is the architecture of how neurons are connected, the activation function is the characteristics of each neuron, and the learning rule is the strategy for learning .ANN is also well suited for parallel implementations because of the simplicity and repetition of the processing elements.
3.2 Supervised Models
Many adaptive, non-parametric neural-net classifiers have been proposed for real-world problems. These classifiers show that they are capable of achieving higher classification accuracy than conventional pixel-based classifiers; however, few neural net classifiers which apply spatial information have been proposed. The feed-forward multilayer neural network has been widely used in supervised image classification of remotely sensed data. A back propagation Feed forward multilayer network is an interconnected network in which neurons are arranged in multilayer and fully connected. There is a value called weight associated with each connection. These weights are adjusted using the backpropagation algorithm or its variations, which is called training the neural networks. Once the network is well trained, it can be used to perform the image classification. Figure (3) shows image that classified with competitive neural network.
4. Minimum Distance Classification
Suppose that each training class is represented by a prototype (or mean) vector:
Where is the number of training pattern vectors from class 
Based on this, we can assign any given pattern
to the class of its closest prototype by
determining its proximity to each 
. If Euclidean distance is our measure of proximity,
then the distance to the prototype is given by
It is not difficult to show that this is equivalent to computing
And assign
to class yields the largest value.
Figure (4) depicts classified image based on minimum distance methods.
abc 
abc1
Figure 3. classified image with neural network Figure 4 .classified image with minimum distance
5. Accuracy assessment
Idea for accuracy assessment methods of classification results comes from the select random sample areas with known classes and then let methods ‘say’ what these samples are. With 100 random selected samples, results were as following:
correctly classified samples in fuzzy logic methods: 72
misclassified: 28
Accuracy: 72%
correctly classified samples in neural network: 82
misclassified: 18
Accuracy: 82%
correctly classified samples in minimum distance: 65
misclassified: 35
Accuracy: 35%
6. Conclusion:
Considering chosen land cover classes, results from image classification and accuracy assessment can be good starting point for certain analysis. Fuzzy logic takes advantage of already created simple rules and image classification equal or even less time consuming. Considering the level of classification accuracy, fuzzy logic can be satisfactory used for image classification. Neural network method has some advantages, stable training results and no requirement of a priori knowledge provided by the simple competitive learning, and optimization for application that need to high precision. As expected neural network method has high precision for classifying image because this method use nonlinear function due to high relative precision .of course fuzzy system is an expert system too. The precision of fuzzy system depends on number and precision of defined rules. This is better that use other spectral bands likes thermal and near infrared bands to increase fuzzy logic rules and due to high precision of classification. In case of minimum distance we should notice that this methods is suitable for image that number of classes be limited and for obtain to high precision if gray levels of different classes are not near to each other . We suggest that for classification of image to obtain high precision expect the geometrical and topological relation between pixels.
6. References
"Supervised classification in high resolution images (Quikbird) using Neural network, Fuzzy sets and Minimum distance"
Haiza Wahida Haron & M. Rafee Majid
Dept. of Urban and Regional Planning
Universiti Teknologi Malaysia
81310 Skudai, Johor, Malaysia
Emails:haiza_net@hotmail.com, rafee@utm.my
Abstract:
This paper compares three main methods of classification in satellite imageries (Quicbird). These classifications are based on fuzzy theory (Mamdani method), neural network and minimum distance. These methods are supervised and implemented in Mat Lab. In this paper an image is divided into three classes including vegetation, urban regions and water land. Only spectral and radiometric characteristic of image pixels are considered and geometrical and topological relation between pixels aren’t used.
1. Introduction
Generally a satellite image contains different features. The goal of image classification is to divide an image into regions which have same characteristics. Manual classification is done with high precision but development of software technology and expert systems promote us to have more precision with high speed and finally less expense .Using a classified image in geographical information system (GIS) results to better interpretation and decision making. In this paper the three main classification approaches including neural network, fuzzy theory and minimum distance are compared.
The rest of the paper is organized as follows. Section 2 describes the concept of pattern classifiers based on fuzzy if–then rules. Section 3 introduces our neural network methods that use for classifying an image. Section 4 then details minimum distance that used Sections 5 shows results of these methods for comparing that which one is better for this image and finally sees the conclusion using of this methods.
2. Fuzzy classification
Fuzzy logic is relatively young theory. Major advantage of this theory is that it allows the natural description, in linguistic terms, of problems that should be solved rather than in terms of relationships between precise numerical values. This advantage, dealing with the complicated systems in simple way, is the main reason why fuzzy logic theory is widely applied in technique. It is also possible to classify the remotely sensed image (as well as any other digital imagery), in such a way that certain land cover classes are clearly represented in the resulting image. In this paper, a priori knowledge about spectral information for certain land cover classes is used in order to classify Quickbird image in fuzzy logic classification procedure.
input (image channels) and output variables (land classes) are introduced in Mat lab’s environment,
Membership functions are defined using results from supervised classification, Mat lab’s Fuzzy Logic Toolbox was used in definition of fuzzy logic inference rules, These rules are tested and verified through the simulation of classification procedure at random sample areas and at the end,
Quick bird image classification was conducted.
2.1 Fuzzy inference system
Fuzzy inference is the process of formulating the mapping from a given input to an output using fuzzy logic. The process of fuzzy inference involves: membership functions, fuzzy logic operators and if-then rules. There are two types of fuzzy inference systems that can be implemented in the Fuzzy Logic Toolbox:
Mamdani-type and
Sugeno-type.
Mamdani's fuzzy inference method is the most commonly seen fuzzy methodology and it expects the output membership functions to be fuzzy sets. After the aggregation process, there is a fuzzy set for each output variable that needs defuzzification. Sugeno-type systems can be used to model any inference system in which the output membership functions are either linear or constant. This fuzzy inference system was introduced in 1985 and also is called Takagi-Sugeno-Kang. Sugeno output membership functions (z, in the following equation) are either linear or constant. A typical rule in a Sugeno fuzzy model has the following form:
If Input 1 = x and Input 2 = y, then Output is z = ax + by + c
For a zero-order Sugeno model, the output level z is a constant (a=b =0). [2]
2.1.1 Membership function
Membership function is the mathematical function which defines the degree of an element's membership in a fuzzy set. The Fuzzy Logic Toolbox in Mat lab’s environment includes 11 built-in membership function types. These functions are built from several basic functions:
Piecewise linear functions,
The Gaussian distribution function,
The sigmoid curve and
Quadratic and cubic polynomial curve
2.1.2 Fuzzy logic operators
The most important thing to realize about fuzzy logical reasoning is the fact that it is a superset of standard Boolean logic. In other words, if the fuzzy values are kept at their extremes of 1 (completely true) and 0 (completely false), standard logical operations will hold. That is, A AND M operator is replaced with minimum - min (A,M) operator, A OR M with maximum - max (A, M) and NOT M with 1-M and others.
2.1.3 If-Then rules
Fuzzy sets and fuzzy operators are the subjects and verbs of fuzzy logic. Usually the knowledge involved in fuzzy reasoning is expressed as rules in the form: If x is A Then y is B where x and y are fuzzy variables and A and B are fuzzy values. The if-part of the rule "x is A" is called the antecedent or premise, while the then-part of the rule "y is B" is called the consequent or conclusion. Statements in the antecedent (or consequent) parts of the rules may well involve fuzzy logical connectives such as ‘AND’ and ‘OR’. In the if-then rule, the word "is" gets used in two entirely different ways depending on whether it appears in the antecedent or the consequent part.
2.2 Classification procedure
The Fuzzy Inference System (FIS) Editor displays general information about a fuzzy inference system: a simple diagram with the names of each input variable (green, red and blue channel) and those of each output variable (water, urban area, vegetation). The Membership Function Editor is used to display and edit all membership functions associated with all of the input and output variables for the entire fuzzy inference system. Figure (1) shows original Quick bird image satellite and figure (2) depicts classified image using fuzzy logic.
Figure1. Quikbird satellite image
Figure 2. classified image with fuzzy logic
3. Neural Network
Artificial neural networks have been employed for many years in many different application areas such as speech recognition and pattern recognition .In general, these models are composed of many nonlinear computational elements operating in parallel and arranged in patterns reminiscent of biological neural nets. Similar to pattern recognition, there exist two types of modes for neural networks – unsupervised and supervised. The unsupervised type of these networks, which possesses the self-organizing property, is called competitive learning networks. A competitive learning provides a way to discover the salient, general features which can be used to classify a set of patterns. Because of the variations of object characteristics, atmosphere condition, and noise, remotely sensed images may be regarded as samples of random processes. Thus, each pixel in the image can be regarded as a random variable. It is extremely difficult to achieve high classification accuracy for most per-pixel classification algorithms (classifiers). Photo interpreters have had pre-eminence in the use of context-dependent information for remote sensing mapping. Neural networks have been recognized as an important tool for constructing membership functions, operations on membership functions, fuzzy inference rules, and other context-dependent entities in fuzzy set theory. In this study, the competitive learning neural networks for the image classification.[10].
3.1 ARTIFICIAL NEURAL NETWORKS FOR MULTISPECTRAL IMAGE CLASSIFIERS
Artificial Neural Networks (ANNs), a brain-style computation model, have been used for many years in different application areas such as vector quantization, speech recognition and image classification. In general, ANN is capable of tolerating the noise, distortion and incompleteness of data taken from the practical applications. Researchers have developed several different paradigms of ANNs. These paradigms are capable of detecting various features represented in input signals. An ANN is usually composed of many nonlinear computational elements. These computational elements operate in parallel to simulate the function of human brain. Hence, an ANN is characterized by the topology, activation function, and learning rules. The topology is the architecture of how neurons are connected, the activation function is the characteristics of each neuron, and the learning rule is the strategy for learning .ANN is also well suited for parallel implementations because of the simplicity and repetition of the processing elements.
3.2 Supervised Models
Many adaptive, non-parametric neural-net classifiers have been proposed for real-world problems. These classifiers show that they are capable of achieving higher classification accuracy than conventional pixel-based classifiers; however, few neural net classifiers which apply spatial information have been proposed. The feed-forward multilayer neural network has been widely used in supervised image classification of remotely sensed data. A back propagation Feed forward multilayer network is an interconnected network in which neurons are arranged in multilayer and fully connected. There is a value called weight associated with each connection. These weights are adjusted using the backpropagation algorithm or its variations, which is called training the neural networks. Once the network is well trained, it can be used to perform the image classification. Figure (3) shows image that classified with competitive neural network.
4. Minimum Distance Classification
Suppose that each training class is represented by a prototype (or mean) vector:
Where
is the number of training pattern vectors from class Based on this, we can assign any given pattern
to the class of its closest prototype by
determining its proximity to each . If Euclidean distance is our measure of proximity,
then the distance to the prototype is given byIt is not difficult to show that this is equivalent to computing
And assign
to class yields the largest value.Figure (4) depicts classified image based on minimum distance methods.
abc abc15. Accuracy assessment
Idea for accuracy assessment methods of classification results comes from the select random sample areas with known classes and then let methods ‘say’ what these samples are. With 100 random selected samples, results were as following:
correctly classified samples in fuzzy logic methods: 72
misclassified: 28
Accuracy: 72%
correctly classified samples in neural network: 82
misclassified: 18
Accuracy: 82%
correctly classified samples in minimum distance: 65
misclassified: 35
Accuracy: 35%
6. Conclusion:
Considering chosen land cover classes, results from image classification and accuracy assessment can be good starting point for certain analysis. Fuzzy logic takes advantage of already created simple rules and image classification equal or even less time consuming. Considering the level of classification accuracy, fuzzy logic can be satisfactory used for image classification. Neural network method has some advantages, stable training results and no requirement of a priori knowledge provided by the simple competitive learning, and optimization for application that need to high precision. As expected neural network method has high precision for classifying image because this method use nonlinear function due to high relative precision .of course fuzzy system is an expert system too. The precision of fuzzy system depends on number and precision of defined rules. This is better that use other spectral bands likes thermal and near infrared bands to increase fuzzy logic rules and due to high precision of classification. In case of minimum distance we should notice that this methods is suitable for image that number of classes be limited and for obtain to high precision if gray levels of different classes are not near to each other . We suggest that for classification of image to obtain high precision expect the geometrical and topological relation between pixels.
6. References
- [1] Abkar A. A., 1999, Likelihood-Based Segmentation and Classification of Remotely Sensed Images, Ph.D thesis, University of ITC, Netherlands
- [2] Nedeljkovic I., Image Classification Based on Fuzzy Logic, MapSoft Ltd, Zahumska 26 11000 Belgrade, Serbia and Montenegro
- [3] Roubos H., M. Setnes and J. Abonyi, Learning Fuzzy Classifcation Rules from Data, Delft University of Technology, Netherlands
- [4] Ünsalan C., K. L. Boyer, 2004, Classifying Land Development in High-Resolution Satellite Imagery Using Hybrid Structural–Multispectral Features, IEEE, Vol. 42, No. 12
- [5] Yanar T. A., Z. Aky¨urek, The Enhancement of ArcGIS with Fuzzy Set Theory
- [6] Marc Bartels and HongWei, Maximum Likelihood Classification of LIDAR Data incorporating multiple co-registered Bands
- [7] Guoliang Fan and Xiang-Gen Xia, Maximum Likelihood Texture Analysis and Classification Using Wavelet-Domain Hidden Markov Models
- [8] Rory Hutson Remote Sensing Group Plymouth Marine Laboratory Remote Sensing and Land Classification
- [9] L. Debnath. Wavelets and Signal Processing. Birkhauser Boston, 2003.
- [10] H. Demuth and M. Beale. Neural Network Toolbox. The MathWorks, Inc., Natick, Massachusetts 01760, 1998.
- [11] http://www.grida.no/prog/global/cgiar/awpack/fuzzy.htm
- [12] http://fuzzy.cs.uni-magdeburg.de/nefclass/
- [13] http://www.iis.sinica.edu.tw/event/html/english/s200702151030.html
- [14]http://homepages.inf.ed.ac.uk/rbf/HIPR2/classify.htm
- [15] http://www.geogr.ku.dk/chips/manual/F159.htm
- [16] http://portal.acm.org/citation.cfm