Analysis of Satellite Images for Historical Maps Classification using Machine Learning Algorithms

Remote sensing is concerned with high-resolution geographical data. Its applications include farming, water resources, ecology, forests, deserts, marine applications, soil types, disaster alerts, and management. To extract the necessary information from satellite images, considerable image processing is required. Image segmentation and categorization into land cover categories is an important stage in satellite image processing. This article describes a method for categorizing and detecting different forms of land cover from satellite data. In this study, we examine satellite images for historical map classification using machine learning algorithms. We acquire Landsat images to complete this tedious task. The images are carefully labeled using the ArcMap software. We categorize areas into five types: forest, building, water, barren, and agricultural land. The images are then sent to the Convolutional Neural Network (CNN) Model for feature extraction. The retrieved features were then exposed to three Machine Learning (ML) models: Support Vector Machine (SVM), Naive Bayes (NB), and Decision Tree (DT). Standard classification measures are used to analyze the outcomes of all three models. When compared to the other two models in satellite image classification, CNN feature extraction with SVM classifier achieves the greatest accuracy of 95%.