Employing Satellite Imagery on Investigation of Convolutional Neural Network Image Processing and Poverty Prediction Using Keras Sequential Model

The primary objective of the proposed model is that the most of the world's poorest individuals reside in areas where national domestic assessments are used to gather info on deficiency. It is difficult to acquire current and precise data because it takes a lot of assets to conduct these surveys. Due to advancements in computer vision and the widespread availability of abundant data sources such as satellite images captured during daylight and nocturnal lighting, a practical solution to the problem of data scarcity is now feasible. This study is going to expand on previous research by processing daytime satellite photos and nighttime lights employing machine learning techniques for the purpose predict, the distribution of poverty at the local level in countries utilizing new technology and modern data sources. Innovative and fascinating possibilities, such as the detailed classification of specific objects on a per-pixel basis, have become feasible due to the availability of aerial satellite data. This study demonstrates the efficacy of a convolutional neural network (CNN) in efficiently and accurately classifying individual pixels inside satellite imagery of a compact urban area. The broad segmentation is then refined by incorporating the expected detailed pixel classifications, enhancing the overall accuracy and speed of the classification process. Examined and assessed are the several architectural decisions made for the CNN architecture. The five different types of terrain, ground cover, roads, structures, and water are all physically categorized and assigned to the study area's land mass. The correctness of classification is compared with other per-pixel classification methods for contrast tests conducted on different terrain areas with a comparable number of categories. Convolutional Neural Networks (CNNs) have demonstrated their efficacy in effectively addressing the task of segmenting and detecting objects in remote sensing data. This is clear from the complete classification and segmentation outcomes achieved by the analysis of a limited number of map segments. The image is categorized into three groups—low, bright, and high—depending on their specific characteristics obtained and luminosity, and the associated wealth index has been forecasted for each.