Comparative Analysis of Various Statistical Learning Algorithms in Classification of Satellite Imagery

Satellite image classification problem is a challenge to the researchers working in this field. Optimized image classification is essential for analyzing the classified images. The process of image classification provides a specific significance comprehension for understanding the specific area of satellite image. In image pre-processing contrast limited adaptive histogram equalization is introduced and for image segmentation active contour methodology is used. This paper focused on linear classifiers in satellite image classification. A comparative analysis is performed on Fisher's Linear Discriminant, Logistic Regression, Naive Bayes Classifier and Perceptron. The area of study is Haridwar city located in the hills area of the state Uttarakhand, India. Input image is classified into six classes like Water class, Wheat class, Eucalyptus Class, RRSand class, dense forest and Grassland class. Two satellites FORMOSAT-2 and LANDSAT-8 image is used in this research. For LANDSAT-8 satellite image Naive Bayes classifier accuracy is higher 82.03
^%
and for FORMOSAT-2 satellite image logistic regression model accuracy is higher 83.15%.