Optimized Neural Architecture for Automatic Landslide Detection from High‐Resolution Airborne Laser Scanning Data

Abstract

An accurate inventory map is a prerequisite for the analysis of landslide susceptibility, hazard, and risk. Field survey, optical remote sensing, and synthetic aperture radar techniques are traditional techniques for landslide detection in tropical regions. However, such techniques are time consuming and costly. In addition, the dense vegetation of tropical forests complicates the generation of an accurate landslide inventory map for these regions. Given its ability to penetrate vegetation cover, high-resolution airborne light detection and ranging (LiDAR) has been used to generate accurate landslide maps. This study proposes the use of recurrent neural networks (RNN) and multi-layer perceptron neural networks (MLP-NN) in landscape detection. These efficient neural architectures require little or no prior knowledge compared with traditional classification methods. The proposed methods were tested in the Cameron Highlands, Malaysia. Segmentation parameters and feature selection were respectively optimized using a supervised approach and correlation-based feature selection. The hyper-parameters of network architecture were defined based on a systematic grid search. The accuracies of the RNN and MLP-NN models in the analysis area were 83.33% and 78.38%, respectively. The accuracies of the RNN and MLP-NN models in the test area were 81.11%, and 74.56%, respectively. These results indicated that the proposed models with optimized hyper-parameters produced the most accurate classification results. LiDAR-derived data, orthophotos, and textural features significantly affected the classification results. Therefore, the results indicated that the proposed methods have the potential to produce accurate and appropriate landslide inventory in tropical regions such as Malaysia.

Sensitivity analysis of automatic landslide mapping: numerical experiments towards the best solution

“…In the case of SVM classification and the same DTM resolution and features, the kappa index is 0.55 and the OA is 81.4%. Similar results have also been demonstrated by other authors (Leshchinsky et al 2015;Mezaal et al 2017). Lower kappa values are likely related to under-prediction of landslides and over-prediction of non-landslide areas.…”

“…In some studies, certain morphometric parameters have been used exclusively to detect landslides (Chen et al 2014;Lin et al 2013b;Van Den Eeckhaut et al 2012) and in other studies, they have sometimes been integrated with other RS data (Chen et al 2017;Kurtz et al 2014;Mezaal et al 2017). McKean and Roering (2004) applied local surface roughness, calculating variability in slope and aspect and the two-dimensional topographic curvature through the Laplacian operator to detect and characterize landslides.…”

“…Recent papers have demonstrated that the combination of many DTM derivatives with sophisticated machine learning classifiers (support vector machine or random forest) is computationally time-consuming (Chen et al 2014;Mezaal et al 2017) and can cause over-fitting issues (Chen et al 2014). The best classification results have been achieved by selecting the most relevant features (Danneels et al 2007;Kursa and Rudnicki 2010;Li et al 2015;Mezaal et al 2017;Pawłuszek and Borkowski 2017a;Stumpf and Kerle 2011).…”

“…Still more recently, Chen et al (2017) applied curvature, hillshade, roughness, flow direction, and slope simultaneously with satellite images and OOA to map a landslide in the Three Gorges Reservoir in China. Mezaal et al (2017) applied 39 features in calculating the grey-level co-occurrence matrix (GLCM) from hillshade, height (normalized digital surface model nDSM), slope, and aspect to optimize a neural network architecture for ALM.…”

Editorial for Special Issue: “Application of Artificial Neural Networks in Geoinformatics”

“…Mezaal, M., Pradhan, B., Sameen, M., Mohd, S. H., Yusoff, Z. [5] used airborne laser scanning images to detect landslides. Chen, G., Li, Y., Sun, G., Zhang, Y.…”