Deep Learning Approach with a Convolutional Neural Network for Landslide Detection Using Remotely Sensed Images

A landslide is a natural hazard that has become prevalent, especially in these times of climate change. Detecting a natural hazard is critical as it can prevent costs and fatalities. In this study, we investigate the efficiency of algorithms in detecting landslides. We implement this investigation using convolutional neural networks to classify extracted features from satellite images to detect landslides. This study investigates the use of convolutional neural networks to classify, segment, detect and landslides in an area. Two experiments were conducted; the first was to evaluate the performance of the deep learning model when the learning rate changes along with the number of epochs, and the second was to evaluate the accuracy of the deep learning model. For experiment 1, performance was better due to a higher epoch number per learning rate by monitoring the loss, and it was then used for experiment 2. Adam optimizer was used to build a deep learning CNN model, and after 20 epochs of training, the model gave an F1-score area of 65.09%, recall area of 81.13%, and precision of 57.44%. The model can detect potential landslide areas from LiDAR image input data.