Comparative analysis of five convolutional neural networks for landslide susceptibility assessment

摘要

To evaluate the performance of deep learning methods on the landslide susceptibility mapping, five different convolutionalneural networks (CNN)-AlexNet, Inception-v3, Xception, ResNet-101, and DenseNet-201-were employed to predict thelandslide susceptibility along a transmission line. Ten landslide influencing factors were extracted from three databases andconsidered in the input layers. The landslide (10,481 grids) and non-landslide (10,481 grids) data were randomly subdividedinto 70 (14,673 grids) and 30 (6289 grids) to construct the training and validation samples, respectively. The appropriatearchitecture and training parameters were carefully selected after many attempts until the training and validation accuracywas above 90. The receiver operating characteristic (ROC) curve, landslide density (LD), and landslide ratio (LR) weredetermined to estimate the five CNN networks’ prediction accuracy. All CNN networks had high area-under-the-curve(AUC) values when assessing landslide susceptibility, and most landslides occurred in the outputs with predicted high andvery high landslide susceptibility (LD > 65 and LR > 2.90). Generally, CNN networks had a higher accuracy than the twotraditional methods due to the powerful capability of deep feature extraction. Additionally, the computational time cost inthree steps was recorded to investigate the efficiency of five CNN networks, and all CNN networks accomplished the goalswithin an acceptable time using a commercially available computer (~ 24 h). Comparatively, AlexNet and Xception hadbetter performance than other networks on the landslide susceptibility assessment.