Abstract: As the universe finds it challenging to define poverty, the world bank views poverty as anyone living below $2 per day. Government and international organizations are working to eradicate this poverty. The study reviews some research on satellite images on the prediction of poverty through the concept of CNN. The study split the datasets satellite images of four (4) countries: Nigeria, Mali, Malawi, and Ethiopia obtained from Kaggle in 90% for training with 15% of it for validation and 10% for testing. The datasets were analyzed using CNN, VGG16, ResNet50, which shows that the VGG16 model performs better than the other two models with the validation accuracy of 94%, while CNN has 91%. ResNet has the lowest validation accuracy of 62%. The rise of high-resolution satellite images that contain extensive data of regions or countries' patterns, features, and landscapes can be applied to determine the economic livelihood of people or nations. The application of satellite images to the prediction of poverty is much easier, faster, and less expensive compared to more the prediction becomes more accessible. This study recommends the need for the availability of large satellites images for every region or country. Future researchers should focus on satellite images to predict poverty and the application of satellite images in detecting crime, road traffic, agricultural soil, and the like.
Keywords: Convolutional Neural Network, Residual Network, Satellite Images, Architectural Neural Network, Support Vector Machine, Visual Geometry Group.
Title: Poverty Prediction by Using Deep Learning on Satellite Images
Author: Sabeer Saeed, Ibrahim Turkoglu
International Journal of Computer Science and Information Technology Research
ISSN 2348-1196 (print), ISSN 2348-120X (online)
Vol. 11, Issue 1, January 2023 - March 2023
Page No: 79-90
Research Publish Journals
Website: www.researchpublish.com
Published Date: 24-March-2023
