FUZZY
LOGIC AND NEURAL NETWORK IN IMAGE RECOGNITION
ADAIKKALAM.A
Information Technology,
Mahendra College of Engineering,
Salem,India.
manojadai92@gmail.com
HARISH.P
Information Technology,
Mahendra college of engineering,
Salem,India.
harash92@gmail.com
ABSTRACT: Neural
network is a massively parallel distributed processing system, made of highly
interconnected neural computing elements that have the ability to learn and
thereby acquire knowledge and make it available for use. Fuzzy logic uses fuzzy
set theory, in which a variable is a member of one or more sets, with a
specified degree of membership .Fuzzy logic is a controller that is varied in
accordance with the input and output situation. In fuzzy logic controllers have
been developed using speed and mechanical power deviations, and a neural
network has been designed to tune the gains of the fuzzy logic controllers. The
network automatically adjusts to a new environment without using any pre
programmed instruction. Fuzzy logic gives the approximate output values. Fuzzy
logic approach is an emerging tool for solving complex problems whose system
behavior is complex in nature. An attractive feature of fuzzy logic control is
its robustness to system parameters and operating conditions’ changes. So
unlike the classical control design, which requires a model for designing the
controller, fuzzy logic incorporates an alternative way which allows one to
design a controller using a higher level of abstraction without knowing the
model. This makes the fuzzy logic controller very attractive for ill-defined
systems or systems with uncertain parameters .so let us see a detailed view of
fuzzy logic neural network in satellite image processing.
Key
words-Fuzzy logic, Artificial Neural Network, Image Processing
1. INTRODUCTION
Uncertainty pervades our everyday life. Uncertainty arises
because of complexity, ignorance and various chances of randomness.
Imprecision, lack of knowledge or from vagueness like the fuzziness inherent in
our natural language. The nature of uncertainty in a problem is a very
important point that engineers should ponder prior to their selection of an
appropriate method to express the uncertainty. Fuzzy sets provide a
mathematical way to represent vagueness in humanistic systems. In the field of
artificial intelligence, neuro-fuzzy refers to combinations of artificial
neural networks and fuzzy logic.
Neuro-fuzzy hybridization results in a hybrid intelligent system that
synergizes these two techniques by combining the human-like reasoning style of
fuzzy systems with the learning and connectionist structure of neural networks.
Neuro-fuzzy hybridization is widely termed as Fuzzy Neural Network (FNN).
Neuro-fuzzy system incorporates the
human-like reasoning style of fuzzy systems through the use of fuzzy sets and a
linguistic model consisting of a set of IF-THEN fuzzy rules. The main strength
of neuro-fuzzy systems is that they are universal approximators with the
ability to solicit interpretable IF-THEN rules. Neural Network and Fuzzy Logic
in Satellite Image Classification. Image classification is the important part
of remote sensing, Image analysis and pattern recognition. Digital Image
Classification is the process of sorting all the pixels in an image into a
finite number of individual classes. Landuse/Landcover classification of
satellite images is an important activity for extracting geospatial information
for military & civil purposes like inaccessible areas. It is difficult to
classify satellites image manually.
. The
images can be classified by probabilistic techniques, maximum likelihood
classifier, parallelepiped etc. but these are very slow and accuracy is very
less. It is not easy to obtain perfect data in real world since most data
contains errors and omissions. To overcome this soft computing techniques which
are based on uncertainty like fuzzy set theory, rough set theory and artificial
neural network are used. The aim of soft computing is to model human
perceptions of the world with inexact expression.
1.1. SATELLITE IMAGING
Remote sensing in the form of aerial photography has been an
important source of land use and land cover information. Image classification
is defined as the extraction of differentiated classes or themes, land use and
land cover categories, from raw remotely sensed digital satellite data. In the
past simple pixel based classifiers
originated which were designed for multi-spectral data. The relationship
between spectral classes and different surface materials or land cover types
may be known beforehand, or can be determined after classification by analysis
of the spectral properties of each class. The spectral pattern of a cell in a
multi-spectral image can be quantified by plotting the raster value from each
band on a separate coordinate axis to locate a point in an imagined “spectral
space”. This spectral space has one dimension for each band in the image. Most
classification methods use some measure of the distance between points in this
spectral space to assess the similarity of spectral patterns. Cells that are
close together in spectral space have similar spectral properties and have a
high likelihood of imaging the same surface features. The aim of soft computing
is to exploit the tolerance for imprecision uncertainty, approximate reasoning
and partial truth to achieve tractability, robustness, low solution cost, and
close resemblance with human like decision making and to find an approximate
solution to an imprecisely/precisely formulated problem.
2. Methods of Image
Classification
Several image classification methods originate from the
remote sensing domain. Various image classification methods are discussed.
2.1 Supervised Method
In a supervised
classification, the analyst identifies in the imagery homogeneous
representative samples of the different surface cover types of interest. These
samples are referred to as training areas. The selection of appropriate
training areas is based on the analyst's familiarity with the geographical area
and their knowledge of the actual surface cover types present in the image. The
analyst is supervising the categorization of a set of specific classes. The
computer uses a special program or algorithm to determine the numerical
"signatures" for each training class. Once the computer has
determined the signatures for each class, each pixel in the image is compared
to these signatures and labelled as the class it most closely resembles
digitally. Thus, in a supervised classification we are first identifying the
information classes which are then used to determine the spectral classes which
represent them. Hence, supervised classifiers require the user to decide which
classes exist in the image, and then to define training areas of these classes.
These training areas are then input into a classification program, which
produces a classified image. Several supervised methods have concentrated on
working with small sample sizes to minimize the manual tasks of domain
scientists to obtain labeled samples.
2.2 Unsupervised Method
Another concept used for image classification is called
unsupervised method. The numerical information in the spectral data classes are
grouped first, and are then matched by the analyst to information classes.
Clustering algorithms are used to determine the natural
groupings or structures in the data. The analyst specifies how many groups or
clusters are to be looked for in the data, the parameters related to the
separation distance among the clusters and the variation within each cluster.
The iterative clustering process may result in some clusters to be combined, or
some clusters may be further broken down by application of the clustering
algorithm. Thus, unsupervised classification is not completely without human
intervention. It does not start with a pre-determined set of classes as in a
supervised classification. Unsupervised classification does not require
training areas, just the number of classes you would like to end up with. This
method lacks efficiency and scalability when larger or multiple images are
needed for classification.
3. Artificial Neural Networks
Artificial neural network has been, used as a powerful tool
for pattern classification. Neural-network classifiers are nonparametric and
therefore may be more robust when distribution is strongly non-Gaussian. During training, the network is
capable of forming arbitrary decision boundaries in the feature space. The
distributed knowledge represented in the neural network and the knowledge got
by training the priori samples of datum makes the neural network complicated. And also its
nonlinear mapping ability to realize the land-cover hierarchical classification
changes from coarse degree to subtle degree. Thus, ANN method generally can get
more high accuracy of the outcome and have been widely used in land-cover/land-use
classification.
However,
it is difficult to train when the data exhibit nonsparse or overlapping pattern
classes which is often the case in practical applications. Neural networks make
no assumption about data distribution and hence have achieved improved image
classification results compared to traditional methods. A neural network is
generally perceived as being a 'black box'.
It is extremely difficult to document how specific
classification decisions are reached. Through network training the knowledge of
image classification can be derived and stored implicitly in numerical forms as
synaptic weights in the network. But these weights have no obvious meaning in
most cases which makes it difficult to interpret these weights due to their
complex nature. As a result we cannot gain any understanding of the problem at
hand due to the lack of an explanatory capability to provide insight into the
characteristics of the dataset. For this reason it impossible to incorporate
human expertise to simplify, accelerate or improve the performance of image
classification, a neural network has to learn from scratch. An explanation
capability should be an integral part of functionality of a trained neural
network in order to make them applicable to complex remote sensing image
classification.
The
advantages of neural networks can thus be summarized as:
•
Resistance to Noise,
•
Tolerance to Distorted Patterns /Images
• Superior
Ability to Recognize Overlapping Pattern
Classes or
Classes with Highly Nonlinear Boundaries
or
Partially Occluded or Degraded Images
•
Potential for Parallel Processing
• Non
parametric
4. Fuzzy Logic
Fuzzy logic has been used in a wide range of problem
domains. A fuzzy set is a set whose elements have degrees of membership. An
element of fuzzy set can be full member or a partial membership value assigned to an element is no longer
restricted to just two values, but can be 0, 1 or any value in between.
Mathematical function which defines the degree of an element’s
membership in a fuzzy set is called membership function. The major advantage of
this theory is the ability to describe the problem naturally in linguistic
terms rather than in terms of relationships between precise numerical values.
Fuzzy systems, on the other hand, have the capability to represent classification
decisions explicitly in the form of fuzzy 'if-then' rules.
Fuzzy sets allow the assignment of partial and multiple valued memberships.
Fuzzy systems make use of vague, imprecise or uncertain information to generate
simpler more suitable
models that are easier to handle and more familiar to human thinking. Human
expert is the main source of fuzzy rules, thus it is possible to improve the
performance of the system by adding new rules, removing defective rules or
update existing rules in the knowledge base. However, the construction of a
knowledge base, especially the fine-tuning of the fuzzy set parameters of the
fuzzy rules in a fuzzy expert system, is a tedious and subjective process.
.
The computer-assisted supervised classification requires sufficiently
homogenous training data to perform the multi-spectral image classification.
Fulfilling this requirement for the image with highly complex surface features
is not feasible. Moreover, the traditional classification mapping with
one-pixel-to-one-class algorithms normally fail to deal with the mixed the
pixels that ordinary caused by the mixture of land cover classes. The complex
land surface often causes the mixed pixels in the remote sensing image if the
image pixel size is not fine enough to catch the spectral response from only a
single land class. For example, a mixed pixel may contain the spectral
responses from both grass and underlying soils. Fuzzy classification has been
used to deal with mixed pixel problem that allows every pixel has a membership
value between 0 and 1 for every candidate class. In the classification of
remote sensing images, Bezdek et al developed a fuzzy c-means clustering
algorithm to perform an unsupervised classification. Wang proposed a supervised
mode for fuzzy classification. Mannan et al applied fuzzy neural networks to
the classification of multi-spectral images. yet all found that the fuzzy membership
values for each cover class strongly correlate with the actual ground
proportions of those land cover classes. In complex fuzzy systems, manual
determination and optimization of fuzzy membership parameters is impossible. It
is desirable that knowledge automation be incorporated into existing fuzzy
systems in order to make the benefits of fuzzy logic available to image
classification.
5. Neuro-Fuzzy Approach
As standalone systems, neural networks and fuzzy logic
exhibit unique features and fundamental limitations. However, it is found that
both the two technologies are complimentary to each other from the functional
point of view. So if the two technologies are combined, one can provide
capabilities not available
in other. The integrated system is called neuro-fuzzy system. A neuro-fuzzy
system is a fuzzy system that uses learning algorithm inspired by neural
network theory to determine
its parameters based on sample data. A neuro-fuzzy system usually delivers more
powerful solutions than its individual components. The learning algorithms of
neural networks developed can be used to automate the derivation of fuzzy set
parameters for the fuzzy 'if-then' rules in a fuzzy expert system. The rules
are in symbolic form and thus facilitate the understanding of the neural network
based image classification system. Also the image classification accuracy
obtained from the improved neuro-fuzzy system was significantly superior to
those of the back propagation based neural network and the maximum likelihood
approaches.
6. Conclusions
Both neural networks and fuzzy logic systems have many
advantages. But the ‘black box’ problem of neural networks and knowledge
automation problem associated with fuzzy systems have hindered the widespread
adoption of these two methods for classification of complex remotely sensed
data. The improved neuro-fuzzy image classification system is based on the
synergism between neural networks and fuzzy expert systems. It incorporates the
best of both technologies and compensates for the shortcomings of each.
No comments:
Post a Comment