Dhristhi - Robust Change Detection, Monitoring, and Alert System on User-defined AOI using Multi-Temporal Sentinel-2 Satellite Imagery

Unregulated and aggressive Land Use Land Cover (LULC) dynamics such as urban sprawl and informal deforestation require a monitoring system that is close to real time, accurate, and readily available. Currently available commercial and cloud-based systems often demand high levels of technical knowledge in geospatial and programming, posing a stumbling block to local practitioners and NGOs. We propose Dhristhi, an automated, end-to-end Web-GIS platform designed to overcome this expertise bottleneck. Dhristhi integrates a sophisticated hybrid methodology: it employs a pre-trained U-Net Deep Learning model for high-precision, pixel-wise semantic segmentation of multispectral imagery, followed by an Object-Based Post-Classification Comparison (OBC) approach to aggregate and validate changes into meaningful geographic regions. An important component of the framework is a Random Forest (RF) classifier with user defined dynamic thresholding mechanism & seasonal variation in values of spectral indices (NDVI, NDBI) to prevent false positive and improve change validation. This platform works on a fully automated, plug-and-play workflow, it automatically handles everything from data collection to pre-processing, so anyone can use the platform easily without having technical expertise. After any successful change detection system automatically sends alert and generates report in the user’s dashboard. Dhristhi successfully demonstrates advanced geospatial intelligence, providing a robust, noise-resilient tool for environmental and civil governance. On a validation dataset covering forest-sprawled and urban regions, the proposed system achieves an overall change detection accuracy of 92.1%, with a Kappa coefficient of about 0.88 and mean IoU of about 0.89 for binary anthropogenic change detection.