A Survey of Various Algorithms Used on Multispectral Satellite Image Classification of Alwar Image Dataset

Objectives: The aim of the survey was to study and compare the efficacy of using bio inspired algorithms for the purpose of classification of a satellite dataset. The final objectives involved identification of a suitable technique to most effectively classify a satellite image using various strategies. Methods/Statistical Analysis: Bio Inspired Algorithms (BIA) are increasingly being used in classification tasks as they are extremely efficient and can generate solutions for complex associations from simple initial settings. The paper presents a survey and review of the use of the BIAs in the domain of satellite image classification to segregate similar pixels with common values for different bands of data and classifying them into various classes. This study concentrates on a multispectral satellite data of size 472 X 576 (257712 pixels) of the Alwar region, Rajasthan having 7 bands of attribute information. The data is classified into classes of major land-use types. Findings: The study has taken into account various strategies used by different researchers to produce the best classification accuracies for classifying a satellite image dataset. This includes the process of feature selection and feature reduction wherein a combination of classifiers are used as ensemblers or in a hybrid model to achieve better classification. The different classification approaches like per pixel, sub pixel, object based and knowledge based classification has been studied and categorized based on their usage. The parameters used to determine the efficacy comparisons for these classifications are namely kappa coefficient, producers, users and overall accuracy. The various classification strategies have been compared based on their kappa coefficient performance. Cuckoo search algorithm and artificial bee colony, two of the recent bio inspired algorithms has shown an impressive classification accuracy and comparable performances are given by the variants of ant colony optimization algorithm in different hybrid models with PSO and BBO. The performances of per pixel and object based classification methodology are similar and the higher accuracies are determined by the use of better classification algorithm. Application/Improvements: In addition certain application usage of the bio inspired classification algorithm in the domain of remote sensing has also been studied. To substantiate the efficient performances of these bio inspired algorithms in the satellite image classification; the effect of these on different sets of satellite images has also been studied and found to be good.