CLASSIFICATION OF LISS IV IMAGERY USING DECISION TREE METHODS

Image classification is a compulsory step in any remote sensing research. Classification uses the spectral information represented by the digital numbers in one or more spectral bands and attempts to classify each individual pixel based on this spectral information. Crop classification is the main concern of remote sensing applications for developing sustainable agriculture system. Vegetation indices computed from satellite images gives a good indication of the presence of vegetation. It is an indicator that describes the greenness, density and health of vegetation. Texture is also an important characteristics which is used to identifying objects or region of interest is an image. This paper illustrate the use of decision tree method to classify the land in to crop land and non-crop land and to classify different crops. In this paper we evaluate the possibility of crop classification using an integrated approach methods based on texture property with different vegetation indices for single date LISS IV sensor 5.8 meter high spatial resolution data. Eleven vegetation indices (NDVI, DVI, GEMI, GNDVI, MSAVI2, NDWI, NG, NR, NNIR, OSAVI and VI green) has been generated using green, red and NIR band and then image is classified using decision tree method. The other approach is used integration of texture feature (mean, variance, kurtosis and skewness) with these vegetation indices. A comparison has been done between these two methods. The results indicate that inclusion of textural feature with vegetation indices can be effectively implemented to produce classified maps with 8.33% higher accuracy for Indian satellite IRS-P6, LISS IV sensor images.


INTRODUCTION
Classification of satellite imagery plays an important role in many application of remote sensing.Classification is a method by which labels or class identifiers are attached to the pixels making up remotely sensed image on the basis of their spectral characteristics.These characteristics are generally measurements of their spectral response in different wavebands.They also include other attributes (e.g.Vegetation indices and Texture).
Spectral vegetation indices in remote sensing have been widely used for the assessment and analysis of the biomass, water, plant and crops (Jackson and Huete, 1991).Vegetation indices (VI) enhances the spectral information and increases the separability of the classes of interest therefore it influences the quality of the information derived from the remotely sensed data.
Texture is also one of the important characteristics used in identifying objects or region in an image.It is an innate property of virtually all surfaces which includes the pattern of different crops in a field.Texture contains important information about the structural arrangement of surfaces and their relationship to surrounding environment.In pixel-based approach, each pixel is classified individually, without considering contextual information.Several studies have explored the potential for using these texture statistics derived from satellite imagery as input features for land cover classification (Haralick et al, 1973, Harris, 1980, Shih et al, 1983).
Many algorithms have been developed and tested to classify satellite images.There are two approaches namely supervised and unsupervised classification, known as hard classifiers.The * Corresponding author traditional hard classification techniques are parametric in nature and they expect data to follow a Gaussian distribution, they have been found to be performing poor results.In order to overcome this problem, non-parametric classification techniques such as artificial neural network (ANN) and Decision tree classification (DT) are used.The non-parametric property means that non homogenous, non-normal and noisy data sets can be handled, as well as non-linear relations between features and classes, missing values, and both numeric and categorical inputs (Quinlan, 1993).Decision tree technique includes a set of binary rules that define meaningful classes to be associated to individual pixels.Different decision tree software are available to generate binary rules.The software takes training set and supplementary data to define effective rules.In this study decision tree approach is used for land cover studies using LISS IV sensor data.

THE STUDY AREA
The selected area for this study is village Foloda which is located in Muzaffarnagar District, India, Measuring approximately 8 km 2 which lies between 29°36'22.70"N-29°38'41.11"NLatitude and 77°47'50.26"E-77°50'38.21"ELongitude.The ground truth information of the study area, including field wise information of various crops and non-crop were collected using Trimble JUNO Global Positing System (GPS).The main crop growing in this region are sorghum, paddy, wheat and sugarcane.The study area is shown in Figure 1.

Generation of Various Vegetation Indices
Various vegetation indices have been developed by linear combination of red, green and near-infrared spectral bands (Basso et al., 2004).Vegetation indices are more sensitive than the individual bands to vegetation parameters (Baret and Guyot, 1991).The eleven vegetation indices Normalized Green (NG), Normalized Red (NR), Normalized Near Infrared (NNIR), Vegetation Index Green (VI green), Difference Vegetation Index (DVI), Normalized Difference Vegetation Index (NDVI), Green Normalized Difference Vegetation Index (GNDVI), Normalized Difference Water Index (NDWI) Optimized Soil Adjusted Vegetation Index (OSAVI), Modified Soil Adjusted Vegetation Spectral Index (MSAVI2) and Global Environmental Monitoring Index (GEMI) are generated using reflectance of green, red and NIR bands of LISS IV sensor.ENVI 5.1 band math function is used for formulation of vegetation indices.The different formulae of vegetation indices are shown by equations ( 1)-( 11) and generated images is shown in Figure 5.
Figure 6.Generated texture images

Decision Tree Classification
The decision tree is an approach where pixels are classified based on a sequence of binary decisions (Safavian and Landgrebe, 1991).According to decision tree, the first conditional statement leads to the second, the second to the third and so on.Decision tree is an inductive learning algorithms which generates classification tree using the training samples.MATLAB 15a was used to build decision trees.In this study training samples are selected based on Google Earth and GPS field observations.The characteristics of training sample ROIs is summarize in

Decision tree classification based on vegetation indices:
The steps used in this classification are given below:

Accuracy Assessment
Accuracy assessment is used to compare the classification results with reference data, which is assumed to be true for determining the classification results.Many methods are used to analyse the accuracy of remotely sensed data (Congalton andGreen, 1999, Koukoulas andBlackburn, 2001).In this work, confusion matrix or error matrix method is used (Foody, 2002).Reference data has been taken during the field visit on September 18-21, 2013.Total 650 pixels have been selected for various classes to determine the accuracy.The accuracy assessment has been done using ERDAS IMAGINE software.The producer's accuracy (PA), user's accuracy (UA), overall accuracy (OA) and kappa coefficient (K) values are given in

RESULTS AND CONCLUSIONS
The final classified images are shown in Figure 7 and Figure 8. Indian satellite IRS-P6 LISS IV sensor imagery has been classified using decision tree method.The first decision tree was constructed based on only vegetation indices and the second one was constructed using vegetation indices with textural features.The final image was classified into eight major classes (water, fallow, settlement, poplar tree, orchard, sugarcane, paddy and sorghum).The overall accuracy and kappa coefficient is found to be 81.08 % and 0.79 for decision tree using vegetation indices method.Inclusion of textural feature with vegetation indices decision tree, overall accuracy and kappa coefficient is 89.42 % and 0.87 respectively.The results indicates that LISS IV imagery can be effectively implemented to produce classified maps with higher accuracy.

Figure 1 .
Figure 1.The study area

Figure
Figure 3. Atmospheric correction SACRS2 model

Table 2 .
Number of ROIs and Pixels in each class