Land-Cover Classification Using Deep Learning with High-Resolution Remote-Sensing Imagery

Land-Cover Classification Using Deep Learning with High-Resolution Remote-Sensing Imagery

2024 | Muhammad Fayaz, Junyoung Nam, L. Minh Dang, Hyoung-Kyu Song, Hyeonjoon Moon
This paper presents a robust approach to land-area classification (LAC) using deep learning with high-resolution remote-sensing imagery. The authors leverage transfer learning with Inception-v3 and DenseNet121 architectures to develop a reliable LAC system. The study addresses the limitations of traditional remote-sensing methods in accurately classifying dynamic and complex urban land areas. By employing transfer learning, the models benefit from pre-trained features on large datasets, enhancing generalization and performance. The fine-tuning process further optimizes the models for specific land-use classes, improving accuracy and reliability. The proposed system achieves impressive results, including 92% accuracy, 93% recall, 92% precision, and a 92% F1-score on the UC-Merced_LandUse dataset. Heatmap analysis provides insights into the decision-making process, highlighting the models' effectiveness in capturing contextual information. The study contributes to the evolution of remote-sensing methodologies and underscores the importance of incorporating advanced techniques like fine-tuning and specific network architectures in LC classification systems.This paper presents a robust approach to land-area classification (LAC) using deep learning with high-resolution remote-sensing imagery. The authors leverage transfer learning with Inception-v3 and DenseNet121 architectures to develop a reliable LAC system. The study addresses the limitations of traditional remote-sensing methods in accurately classifying dynamic and complex urban land areas. By employing transfer learning, the models benefit from pre-trained features on large datasets, enhancing generalization and performance. The fine-tuning process further optimizes the models for specific land-use classes, improving accuracy and reliability. The proposed system achieves impressive results, including 92% accuracy, 93% recall, 92% precision, and a 92% F1-score on the UC-Merced_LandUse dataset. Heatmap analysis provides insights into the decision-making process, highlighting the models' effectiveness in capturing contextual information. The study contributes to the evolution of remote-sensing methodologies and underscores the importance of incorporating advanced techniques like fine-tuning and specific network architectures in LC classification systems.
Reach us at info@study.space