Using Satellite Imagery to Map Poverty Struck Areas in Pakistan Using Neural Networks

EasyChair Preprint 14195

Abstract

Poverty is a complicated socioeconomic issue that comprises of more than just financial difficulty. It also includes inadequate healthcare, education, and basic housing. Satellite imaging has become a powerful tool for studying socioeconomic trends, especially in the areas where poverty is a problem. Making use of this potential, this research aims to provide a solid model for mapping poverty in Pakistan, which would facilitate resource allocation and decision-making. The absence of reliable data makes it difficult to accurately measure Pakistan's poverty levels, even with advances in satellite technology. Satellite imagery-trained convolutional neural networks (CNNs & ANNs) are becoming one of the most widely used and successful methods. ResNet50, ResNet101 and Yolo are deep convolutional neural network (CNN) architectures well-known for their ability to train deeper networks effectively by utilizing residual blocks. Our goal is to increase efficiency and performance of the system by using ResNet and Yolo models for both training and efficiency comparison. These methods represent a significant advancement in satellite image processing and provide enhanced capabilities for resource allocation in poverty reduction programs and evidence-based decision-making. These maps facilitate policymakers, researchers, and NGOs by providing insightful information. These results shows that the proposed scheme is effective in creating poverty maps with high accuracy (75.5%) and precision (76.2).

Keyphrases: Accuracy and Precision in Poverty Assessment, Convolutional Neural Networks (CNNs), Poverty Reduction Programs, ResNet50, Resnet101, Socioeconomic Analysis, evidence-based decision making, poverty mapping, satellite imagery

@booklet{EasyChair:14195,
  author    = {Hamza Umer Farooq and Muhammad Junaid Nazar and Muhammad Sarib Hanif and Aamna Faisal},
  title     = {Using Satellite Imagery to Map Poverty Struck Areas in Pakistan Using Neural Networks},
  howpublished = {EasyChair Preprint 14195},
  year      = {EasyChair, 2024}}