Land Classification Using Satellite Images Through LSSVM Algorithm

Abstract

Land classification plays a crucial role in environmental monitoring and resource management, utilizing satellite imagery to delineate distinct land cover types. In this study, we explore a comprehensive approach to land classification, employing a sequence of preprocessing, segmentation, and classification algorithms. Discrete Wavelet Transform (DWT) and Stationary Wavelet Transform (SWT) are employed for image preprocessing, extracting valuable spatial information from IKONOS and Sentinel satellite images. Subsequently, K-means clustering, Particle Swarm Optimization (PSO), Discrete Particle Swarm Optimization (DPSO), and Fractional Order Discrete Particle Swarm Optimization (FODPSO) are utilized for segmentation, effectively delineating land cover boundaries.Furthermore, a novel contribution is introduced by proposing the use of Least Squares Support Vector Machine (LSSVM) as the classification algorithm. LSSVM is demonstrated to outperform other algorithms in terms of precision, recall, accuracy, and F1-Score. Specifically, LSSVM exhibits a remarkable accuracy of 96%, surpassing the performance of DWT, SWT, K-means clustering, PSO, DPSO, and FODPSO on both IKONOS and Sentinel satellite images. This substantiates the efficacy of the proposed LSSVM-based approach in achieving high-precision land classification.The findings underscore the significance of integrating preprocessing, segmentation, and classification techniques for accurate and robust land classification. The proposed LSSVM algorithm stands out as a promising solution for achieving superior accuracy, paving the way for enhanced applications in environmental monitoring and land resource management.