|
Building Change Detection in High Resolution Satellite Images: Comparison of Classical and Deep Learning Methods
|
Mohaddeseh Mesvari   , Reza Shah-Hosseini * , Mohsen Niroomand |
|
|
|
Abstract: (11 Views) |
Detecting changes in urban structures using satellite imagery plays a key role in urban planning, infrastructure development, decision-making, and disaster management. The rapid growth of cities, combined with natural disasters such as earthquakes and floods, has led to significant transformations in urban areas, highlighting the need for accurate and ongoing monitoring. Traditional change detection methods based on manual image analysis face challenges such as high costs, limited scalability, and time-consuming processes. Recent advances in image processing and deep learning have enabled automated detection of structural changes. This study evaluates and compares several methods, including traditional approaches like Change Vector Analysis (CVA), Multivariate Alteration Detection (MAD), and Slow Feature Analysis (SFA), as well as deep learning-based techniques such as Deep Slow Feature Analysis (DSFA) and Dual-Attention Networks (DAS). While CVA and MAD are effective for identifying major changes, they perform poorly in detecting subtle or complex transformations. For example, CVA and MAD achieved accuracies of 49.32% and 65.67%, respectively. In contrast, DSFA reached 84.07% accuracy, although its high computational demands limit its practicality. The DAS method demonstrated the best performance, achieving 95.68% accuracy by effectively using spatial and spectral attention mechanisms to distinguish real changes from noise. The results highlight the advantages of deep learning techniques in providing accurate, efficient, and automated change detection, offering valuable support for urban monitoring and decision-making. |
Article number: 4 |
| Keywords: Change Detection, Satellite Imagery, Deep Learning Methods, Deep Slow Feature Analysis, Dual-Attention Networks. |
|
|
|
Type of Study: Research |
Subject:
Photo&RS
Received: 2024/12/11 | Accepted: 2026/02/7
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
