Review on remote sensing methods for landslide detection using machine and deep learning
Amrita Mohan,Amit Kumar Singh,Basant Kumar,Ramji Dwivedi
DOI: https://doi.org/10.1002/ett.3998
IF: 3.6
2020-06-23
Transactions on Emerging Telecommunications Technologies
Abstract:<p>Landslide, one of the most critical natural hazards, is caused due to specific compositional slope movement. In the past decades, due to inflation of urbanized area and climate change, a compelling expansion in landslide prevalence took place which is also termed as mass/slope movement and mass wasting, causing extensive collapse around the world. The principal reason for its pursuance is a reduction in the internal resistance of soil and rocks, classified as a slide, topple, fall, and flow. Slopes can be differentiated based on earth material and the nature of its movements. The downward flow of landslides occurs due to excessive rainfall, snowmelt, earthquake, volcanic eruption, and so on. This review article revisits the conventional approaches for identification of landslides, predicting future risk, associated with slope failures, followed by emphasizing the advantages of modern geospatial techniques such as aerial photogrammetry, satellite remote sensing images (ie, panchromatic, multispectral, radar images), Terrestrial laser scanning, and High‐Resolution Digital Elevation Model (HR‐DEM) in updating landslide inventory maps. Machine learning techniques like Support Vector Machine, Artificial neural network, deep learning has been extensively used with geographical data producing effective results for assessment of natural hazard/resources and environmental research. Based on recent studies, deep learning is a reliable tool addressing remote sensing challenges such as trade‐off in imaging system producing poor quality investigation, in addition, to expedite consequent task such as image recognition, object detection, classification, and so on. Conventional methods, like pixel and object‐based machine learning methods, have been broadly explored. Advanced development in deep learning technique like CNN (Convolutional neural network) has been extensively successful in information extraction from an image and has exceeded other traditional approaches. Over the past few years, minor attempts have been made for landslide susceptibility mapping using CNN. In addition, small sample sizes for training purpose will be major drawback and notably remarkable while using deep learning techniques. Also, assessment of the model's performance with diverse training and testing proportion other than commonly utilized ratio, that is, 70/30 needs to be explored further. The review article briefly highlights the remote sensing methods for landslide detection using machine learning and deep learning.</p>
telecommunications