1.1 Gaussian Maximum Likelihood Classifier (GMLC) : The maximum likelihood classifier quantitatively evaluates both the variance and covariance of the category spectral response patterns when classifying an unknown pixel. It is assumed that the distribution of the cloud of points forming the category training data is Gaussian (normally distributed). Here, the distribution of a category response pattern can be completely described by the mean vector and the covariance matrix. The probability density functions are used to classify an unidentified pixel by computing the probability of the pixel value belonging to each category. After evaluating the probability in each category, the pixel is assigned to the most likely class (highest probability value) or can be labelled as ‘unknown' if the probability values are all below a threshold set by the analyst [4] .

1.2 Spectral Angle Mapper (SAM) : In N dimensional multi-(or hyper-) spectral space a pixel vector x has both magnitude (length) and an angle measured with respect to the axes that defines the coordinate system of the space [5]. In the Spectral Angle Mapper (SAM) technique for identifying pixel spectra only the angular information is used. SAM is based on the idea that an observed reflectance spectrum can be considered as a vector in a multidimensional space, where the number of dimensions equals the number of spectral bands. If the overall illumination increases or decreases (due to the presence of a mix of sunlight and shadows), the length of this vector will increase or decrease, but its angular orientation will remain constant. Smaller angles represent closer matches to the reference spectrum. If this angle is smaller than a given tolerance level, the spectra are considered to match even if one spectrum is much brighter than the other (farther from the origin) overall [4]. Pixels further away than the specified maximum angle threshold are not classified.

1.3 Neural Network : To overcome difficulties in conventional digital classification that uses the spectral characteristics of the pixel as the sole parameter in deciding to which class a pixel belongs to, new approaches such as Neural Networks (NN) are being used. Fully trained, neural networks can perform image classification relatively rapidly, although the training process itself can be quite time consuming. NN systems are ‘self-training' in that they adaptively construct linkages between a given pattern of input data and particular outputs. A NN consists of a set of three or more layers, each made up of multiple nodes. Typically, these might include spectral bands from a remotely sensed image, textural features or other intermediate products derived from such images, or ancillary data describing the region to be analysed. The nodes in the output layer represent the range of possible output categories to be produced by the network [4]. Between the input and output layers are one or more hidden layers. These consist of multiple nodes, each linked to many nodes in the preceding layer and to many nodes in the following layer. These linkages between nodes are represented by weights, which guide the flow of information through the network. The number of hidden layers used in a neural network is arbitrary. An increase in the number of hidden layers permits the network to be used for more complex problems but reduces the network's ability to generalise and increases the time required for training . Applying a NN to image classification makes use of an iterative training procedure in which the network is provided with matching sets of input and output data. Each set of input data represents an example of a pattern to be learned, and each corresponding set of output data represents the desired output that should be produced in response to the input. During the training process the network autonomously modifies the weights on the linkages between each pair of nodes in such a way as to reduce the discrepancy between the desired output and the actual output [4].

1.4 Decision Tree Approach : Decision tree approach is a non-parametric classifier and an example of machine learning algorithm. It involves a recursive partitioning of the feature space, based on a set of rules that are learned by an analysis of the training set. A tree structure is developed where at each branching a specific decision rule is implemented, which may involve one or more combinations of the attribute inputs. A new input vector then ‘travels' from the root node down through successive branches until it is placed in a specific class . The thresholds used for each class decision are chosen using minimum entropy or minimum error measures. It is based on using the minimum number of bits to describe each decision at a node in the tree based on the frequency of each class at the node. With minimum entropy, the stopping criterion is based on the amount of information gained by a rule (the gain ratio) [2].

1.5 Clustering : Clustering techniques fall into a group of undirected data mining tools [6]. The goal of clustering is to discover structure in the data as a whole. There is no target variable to be predicted and thus no distinction is being made between independent and dependent variables. Clustering partitions the image data into a number of spectral classes, and then labels all pixels of interest as belonging to one of those spectral classes, although the labels are purely nominal (e.g. A, B, C, …., or class1, class 2, …….) and are as yet unrelated to ground cover types [5] . The K-means algorithm is a simple, iterative procedure, in which a crucial concept is the one of ‘ centroid' . Centroid is an artificial point in the space of records which represents an average location of the particular cluster. The coordinates of this point are averages of attribute values of all examples that belong to the cluster.