Status: published
Satellite remote sensing is widely used for monitoring land use and land cover changes such as urban growth, agriculture expansion, deforestation, and water body variation. However, manual analysis of satellite images is slow, difficult, and not suitable for large-scale monitoring. To solve this problem, this project presents an AI-based satellite image segmentation and change detection system. In this work, a pretrained deep learning segmentation model from Hugging Face was first tested, but its accuracy was low due to outdated training data. Therefore, the model was fine-tuned using a satellite image segmentation dataset collected from Kaggle. The dataset contains real satellite images along with their corresponding ground truth mask images. The trained model classifies each pixel of a satellite image into one of seven land cover classes: urban land, agriculture land, rangeland, forest land, water, barren land, and unknown. After training, the model produces accurate colour-coded segmentation maps. The system also performs change detection by comparing segmentation results of satellite images taken at different time periods. This helps in identifying changes such as urban expansion, deforestation, agricultural growth, and water body variation. The proposed system provides an efficient and automatic solution for satellite image analysis and can be useful for applications such as urban planning, environmental monitoring, and disaster management.
Keywords: Satellite Image Segmentation, Deep Learning, Land Cover Classification, Change Detection, Remote Sensing, Semantic Segmentation, Urban Monitoring, Environmental Analysis.