Landslide Detection and Early Warning System Using Extended Kalman Filter with Convolutional Neural Networks Based Long Short-Term Memory

Nowadays, landslides are a major risk to human life and causing devastating consequences. and economic stability by resulting substantial losses annually. The effective landslide risk management relies on timely recognition of spatial changes in patterns. patterns. Traditional approaches for developing the warning system for landslide had faced challenges which includes linearity assumption and difficulty in handling non-stationary data. Therefore, this paper proposes Extended Kalman Filter-Convolutional Neural Networks-Long Short-Term Memory (EKF-CNN-LSTM) for landslide detection and early warning system. Initially, the data is collected from Fast Deploying Monitoring System (FDMS) which consists of ad-hoc router, sensors related to landslides. Then, the data is preprocessed by using Fast Fourier Transform (FFT), which effectively filtered out noise and irrelevant frequency components by improving signal quality. After that, model is trained by proposed EKF-CNN-LSTM which generated real time prediction, early warnings of potential landslides. Finally, the early warnings are predicted and give output as “no warning” or “warning,” by enabling the timely interventions. By integrating these technologies into landslide monitoring and EWS, enhanced the prediction accuracy and provided timely warnings. The proposed EKF-CNN-LSTM succeeded with high results including accuracy (0.943), precision (0.832), recall (0.786), F1-score (0.798) and MCC (0.743) when compared with existing ResNet101.