TRACKING OF HUMAN ACTIVITIES

Environmental changes driven by human activities—such as agricultural conversion, deforestation, urban expansion, mining, and—are increasing at an accelerated rate, making continuous monitoring essential for sustainable environmental management. Satellite imagery provides a reliable source for observing these changes over time, but manual interpretation is slow and prone to inconsistencies. This paper will presents an automated framework for detecting human-induced environmental changes using multi-temporal satellite images and a Seamless U-Net model tailored for pixel- level change segmentation. The proposed architecture processes paired satellite images from different time periods and learns discriminative spatiotemporal features to accurately distinguish change and no-change regions. A preprocessing pipeline involving radiometric normalization, patch extraction, and augmentation is integrated to improve model robustness under varying imaging conditions. Experimental evaluation is conducted using publicly available satellite datasets containing real examples of anthropogenic change. Results demonstrate that the Seamless U-Net achieves higher IoU, F1-Score, and boundary accuracy compared to standard U-Net variants and traditional change- detection approaches. The model produces cleaner change maps with fewer false alarms, especially in heterogeneous land-cover regions. The findings highlight the potential of deep learning-based bi-temporal analysis for scalable environmental change monitoring and decision-support applications.