Satellite Based Geographical Poverty Prediction

Poverty in the world remains more challenging than traditional methods of assessment such as survey data. It will be logistically difficult to perform the surveys, where they need to assign a few people to perform surveys and to store, backup, and maintain the data without any hurdles. During the survey, the surveyors might miss some of the regions or houses during the survey which might lead to inaccurate results because there may be situations where those regions or houses might be the key factors to determine the poverty and performing these surveys might be time-consuming. Our project mainly focuses on the usage of satellite images to find the solution for the problem faced using the ground survey data where the model is trained using satellite images such as daylight images. The model is trained using CNN transform learning algorithms such as ResUnet-A and UNet algorithms which help in classification by considering the following features such as buildings, roads, water resources, vegetation, and unpaved lands these features will be extracted. These features are key factors in identifying whether the region will be classified into above-poverty, average-poverty, or below-poverty classes. The model helps in providing scalable and cost-effective solutions for poverty assessment and this helps in eliminating the need for ground-level survey data. By providing insights about the images and poverty level it helps the policymakers to allocate the resources based on the insights provided by the model and the use of satellite imagery ensures regular updates, allowing for real-time. This research aims to establish a reliable and accurate poverty prediction contributing to global efforts in reducing the poverty and equitable resource distribution. This research study not only addresses the limitation of the traditional method but also makes us understand the different ways to use the remote sensing images effectively.