A Spatio-Temporal Hybrid Intelligence System for Unauthorized Land Use Detection and Flood Impact Quantification from Satellite Data

The paper describes a hybrid deep learning based system to detect illegal land occupation and estimate the extent of flood damages through the use of satellite image. The model is a composite of VGG16 and Long Short-Term Memory (LSTM) networks used to extract spatial and analyze temporal changes over multi-temporal satellite data respectively. VGG16 model learns the spatial patterns at detail level whereas LSTM learns the time-sequence variations in order to detect unauthorized encroachment of the land and flood prone areas. A fuzzy logic module is used to determine changing degrees of a flood damage using derived features to overcome challenges encountered in satellite observations. The system handles both pre- and post-event imagery and creates automated notifications to authorities to assist in the fast process of making decisions. Data handling and preprocessing are done using geospatial libraries GDAL, OpenCV. The results of the experimental studies show that the land-use change is high precision of land-use change detection and dependability in flood damage assessment, which implies the effectiveness of the framework in real-time tracking and sustainable land management.