부산대학교 도서관

This work aims at depicting a more sophisticated deep learning system for satellite images analysis and further land use categorization that is responsive to various environmental conditions. Therefore, the innovative approach under consideration is based on CNNs trained with the help of multi-spectral and multi-temporal data and focused on transfer learning and data fusion methodologies. The framework is then evaluated against baseline models for its accuracy, precision, recall, and F fine measure and result in substantial gains in all of them depending on the land use classes and environmental conditions. Of particular importance in the proposed approach is the use of Transfer Learning and integration of Multi-modal data. The evaluation shows that the proposed model is superior to this simple strategy and forms powerful method for various applications, including urban planning, environment controlling, and resource administration. The study shows the necessity of the application of progressive deep learning methods to solve the problem in satellite imagery analysis for further development of efficient and improved remote sensing technology.