DEEP LEARNING MODELING OF MULTISPECTRAL SATTELITE DATA FOR LAND USE AND LAND COVER CLASSIFICATION (LULC)

ABSTRACT

This study aims to develop an efficient model for land use and land cover classification (LULC) in Edo state, Nigeria, by employing the pattern recognition abilities of Artificial Neural Networks (ANNs). The research seek to address the vital task of land cover classification which is essential to urban planning, natural resource management, mineral exploration, environmental monitoring, agriculture and conversation. This research utilizes multispectral surface reflectance imagery from Sentinnel-2 as the main data source, Google Earth Engine (GEE) for data collection/preparation. The data used was curated using Google Earth Engine and extensive data preprocessing was performed to ensure a high-quality dataset was used for training the model. Google’s TensorFlow machine learning library was employed for creating and training the Machine Learning model. Leveraging the capabilities of Artificial Neural Networks, a deep learning model was created to extract complex features from the surface reflectance data to ensure accurate classification of land cover types. The model was trained with ground truth data. The results demonstrate a high level of accuracy of 68 per cent, demonstrating the potential of automated land classification systems for applications in land management, conservation, disaster monitoring, scientific research, and urban planning. This thesis contributes to the evolving field of geospatial analysis and remote sensing offering an improvement on traditional methods of land use and land cover classifications by utilizing the synergy of deep learning models and multispectral imagery.