High-Resolution Remote Sensing Satellite Images Classification and Retrieval Model based on Gray Level Co-Occurrence Matrix

Abstract

Natural disasters, deforestation and desertification, forest fires, illicit tree felling for agriculture, urban expansion, and climate change monitoring are all aided by remote sensing satellite images. The volume of image datasets is evolving exponentially as remote-sensing technology advances and the number of Earth observation satellites rises. Machine learning algorithms are capable of quickly and efficiently classifying and retrieving images. Image pattern identification and classification process are done by different Machine learning algorithms from an image search engine that is given by input query images. The Naïve Bayes, SVM Linear, Decision tree, and Random Forest algorithms accuracy are evaluated here. The capability of making effective classification and quick predictions are supported by Naive Bayes Classifier with rapid machine learning models. Both SVM linear and non-linear algorithm achieves high accuracy while utilizing less computational power. The decision tree is the most powerful and widely used algorithm for image categorization and prediction. To improve the dataset's forecast accuracy, Random Forest aggregates the outcomes of numerous decision trees applied to different subsets of the dataset. This study proves that the accuracy of Naive Bayes Classifier is 60%, SVM of 61%, Decision Tree 62%, and Random Forest 65% when using the UC Merced dataset.