Landslide Detection Using Residual Networks and the Fusion of Spectral and Topographic Information

Abstract

Landslide inventories are in high demand for risk assessment of this natural hazard, particularly in tropical mountainous regions. This research designed residual networks for landslide detection using spectral (RGB bands) and topographic information (altitude, slope, aspect, curvature). Recent studies indicate that deep learning methods such as convolutional neural networks (CNN) improve landslide mapping results compared to traditional machine learning. But the effects of network architecture designs and data fusion remain largely underexplored in landslide detection. We compared a one-layer CNN with two of its deeper counterparts and residual networks with two fusion strategies (layer stacking and feature-level fusion) to detect landslides in Cameron Highlands, Malaysia. Sixteen different maps were created using proposed methods and evaluated in separate training and testing sub-areas based on overall accuracy, F1-score, and mean intersection over union (mIOU) metrics. When layer stacking is used as a fusion approach, none of the network designs improved landslide detection results. However, our findings showed that when using feature-level fusion, results could be enhanced with the same network designs. Residual networks performed best improving F1-score and mIOU by 0.13 and 12.96%, respectively, using feature-level fusion rather than layer stacking. CNN models also enhanced the detection outcome with the same fusion approach. On single modality datasets, models' performance varies according to input data, highlighting the effects of input data on network architecture selection. In general, residual networks found to converge faster and generalize better to test areas than other models tested in this research.

Mapping Landslides on EO Data: Performance of Deep Learning Models vs. Traditional Machine Learning Models

“…The popularity of deep-learning to map landslides from EO images is increasing rapidly [29,40]. Here we approached landslide mapping using CNN as a semantic segmentation task, which was lacking in previous works.…”

“…The features maps learned in the intermediate layers can also identify past displacement related geomorphic features, which were very difficult to identify using classical satellite image processing methods. Deep-learning is an emerging method in the field of mapping landslides, and to the best of our knowledge, only a few studies have implemented CNN based landslide mapping [29,39,40,60]. Unlike pixel-based and object-based methods, a CNN can directly learn from images, which removes the need for sampling information in the form of numeric FVs.…”

“…They observed the results of CNN to be comparable with conventional machine learning methods. Sameen and Pradhan [40] have trained a residual networks (ResNet) on spectral and topographic features to map landslide inventory while comparing different feature merging strategies. However, using semantic segmentation based deep-learning architecture (like U-Net) is expected to outperform a sliding-window CNN for detection of landslides [45].…”

Anti-Injury Function of Complex Spiking Neural Networks Under Random Attack and Its Mechanism Analysis

“…Drawing from the advantage of the biological brain, it is significant to research the robustness of artificial neural network (ANN) based on brain-like intelligence. Spiking neural network (SNN) is the most biologically interpreted ANN and has been applied in many fields, such as speech recognition [2], geological monitoring [3], and tumor detection [4]. There are three elements of constructing a SNN: neuron model, synaptic plasticity model, and network topology.…”

Machine Learning for Landslides Prevention: A Survey

“…Then objects are classified (e)[86] (d)[78]. In deep learning, CNN autonomously extracts the contextual features of an image dataset and learn to identify landslide features (f)[87] by looking at R, G, B channels (g)[69].…”