Satellite Images Classification using MobileNet V-2 Algorithm

Abstract

Satellite imagery is an invaluable source of visual information for environmental monitoring and land mapping with high resolution and wide coverage. In this modern technological era, advances in Deep Learning technology have brought great benefits in utilizing satellite images for various purposes. One of the efficient Deep Learning models for satellite image classification is MobileNet V-2, which is specifically designed for devices with limited resources such as smartphones. This study aims to develop an accurate satellite image classification model using Convolutional Neural Network algorithm and MobileNet V-2 model. The data used is taken from the RSI-CB256 dataset developed through crowdsourcing data. This research resulted in the performance of three deep learning models, namely ResNet50, MobileNet V-2, and VGG-16. ResNet50 is the highest model performed best during the training phase, achieve an accuracy of 98.40%. MobileNet V-2 and VGG-16 followed with 95.64% and 96.62% accuracy, respectively. The evaluation results demonstrate the model's strong ability to accurately classify satellite imagery and strengthen the model's ability to generalize well. With high accuracy and the ability to run on smartphone devices, this model has the potential to provide valuable information for governments and scientists in preserving the earth and better responding to environmental changes.