A Feature Engineering Framework for High-Precision Land Use and Land Cover Classification Using Remote Sensing Images

Abstract

Accurate Land Use/Land Cover (LULC) classification is important for efficient natural resource management and for understanding how weather conditions modify the environment. For precise and effective LULC classification, researchers need higher-quality databases and robust classification models. As classification models, geospatial analytical tools, and satellite data become more accessible, it is vital to systematically evaluate different combinations of satellite data and classification methods to help select the best approach for LULC classification. Currently, Deep Learning (DL) algorithms have gained popularity and have become the new standard for data handling in remote sensing research. DL architectures have drawn significant attention owing to their superior performance in classification tasks and other machine vision applications. DL-based classification of very high-resolution satellite images is an important task that provides valuable information for various geospatial applications, particularly LULC mapping. In this paper, we present a novel Enhanced Object Detection and Feature Engineering for High-Precision Land Use and Land Cover (EODFE-HPLULC) algorithm for satellite remote sensing. The main purpose of the EODFE-HPLULC methodology is to demonstrate its efficacy in achieving high-precision classification for real-world remote sensing applications. For object detection, YOLOv12 can be deployed to provide fast, accurate identification of land features. Next, the EfficientNet-B5 model serves as the backbone for feature engineering, extracting high-level, discriminative features from the detected objects. Finally, the deep belief network model performs LULC classification. To evaluate the enriched classification results of the EODFE-HPLULC model, we conduct a comprehensive set of experiments on the benchmark dataset. The comprehensive comparison results demonstrated that the EODFE-HPLULC model outperformed recent techniques, achieving a classification accuracy of 99.14% on the Remote Sensing Satellite Image dataset.