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.

No abstract

No abstract

Abstract:In this paper, a deep learning model using a Recurrent Neural Network (RNN) was developed and employed to predict the injury severity of traffic accidents based on 1130 accident records that have occurred on the North-South Expressway (NSE), Malaysia over a six-year period from 2009 to 2015. Compared to traditional Neural Networks (NNs), the RNN method is more effective for sequential data, and is expected to capture temporal correlations among the traffic accident records. Several network architectures and configurations were tested through a systematic grid search to determine an optimal network for predicting the injury severity of traffic accidents. The selected network architecture comprised of a Long-Short Term Memory (LSTM) layer, two fully-connected (dense) layers and a Softmax layer. Next, to avoid over-fitting, the dropout technique with a probability of 0.3 was applied. Further, the network was trained with a Stochastic Gradient Descent (SGD) algorithm (learning rate = 0.01) in the Tensorflow framework. A sensitivity analysis of the RNN model was further conducted to determine these factors' impact on injury severity outcomes. Also, the proposed RNN model was compared with Multilayer Perceptron (MLP) and Bayesian Logistic Regression (BLR) models to understand its advantages and limitations. The results of the comparative analyses showed that the RNN model outperformed the MLP and BLR models. The validation accuracy of the RNN model was 71.77%, whereas the MLP and BLR models achieved 65.48% and 58.30% respectively. The findings of this study indicate that the RNN model, in deep learning frameworks, can be a promising tool for predicting the injury severity of traffic accidents.

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.