Landslide prediction with severity analysis using efficient computer vision and soft computing algorithms

Since the preceding decade, there has been a great deal of interest in forecasting landslides using remote-sensing images. Early detection of possible landslide zones will help to save lives and money. However, this approach presents several obstacles. Computer vision systems must be carefully built since normal image processing does not apply to images obtained by remote sensing (RS). This research proposes a novel landslide prediction method with a severity analysis model based on real-time hyperspectral RS images. The proposed model consists of phases of pre-processing, dynamic segmentation, hybrid feature extraction, landslide prediction, and landslide severity detection. The pre-processing step performs the geometric correction of input RS images to suppress the built-up regions, water, and vegetation using the Normal Difference Vegetation Index (NDVI). The pre-processing stage encompasses many steps, including atmospheric adjustments, geometric corrections, and the elimination of superfluous regions by denoising techniques such as 2D median filtering. Dynamic segmentation is employed to segment the pre-processed picture for Region of Interest (ROI) localization. The ROI image is utilized to extract manually designed features that accurately depict spatial and temporal variations within the input RS image. For each input RS image, the hybrid feature vector is normalized. We trained ANN and SVM to predict landslides. If the input image predicts a landslide, its severity is identified. For the performance analysis, we collected real-time RS images of the western region of India (Goa and Maharashtra). Simulation results show the efficiency of the proposed model. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.