A comprehensive transferability evaluation of U-Net and ResU-Net for landslide detection from Sentinel-2 data (case study areas from Taiwan, China, and Japan)

Abstract

Earthquakes and heavy rainfalls are the two leading causes of landslides around the world. Since they often occur across large areas, landslide detection requires rapid and reliable automatic detection approaches. Currently, deep learning (DL) approaches, especially different convolutional neural network and fully convolutional network (FCN) algorithms, are reliably achieving cutting-edge accuracies in automatic landslide detection. However, these successful applications of various DL approaches have thus far been based on very high resolution satellite images (e.g., GeoEye and WorldView), making it easier to achieve such high detection performances. In this study, we use freely available Sentinel-2 data and ALOS digital elevation model to investigate the application of two well-known FCN algorithms, namely the U-Net and residual U-Net (or so-called ResU-Net), for landslide detection. To our knowledge, this is the first application of FCN for landslide detection only from freely available data. We adapt the algorithms to the specific aim of landslide detection, then train and test with data from three different case study areas located in Western Taitung County (Taiwan), Shuzheng Valley (China), and Eastern Iburi (Japan). We characterize three different window size sample patches to train the algorithms. Our results also contain a comprehensive transferability assessment achieved through different training and testing scenarios in the three case studies. The highest f1-score value of 73.32% was obtained by ResU-Net, trained with a dataset from Japan, and tested on China’s holdout testing area using the sample patch size of 64 × 64 pixels.

Landslide Detection Based on ResU-Net with Transformer and CBAM Embedded: Two Examples with Geologically Different Environments

“…ResU-Net [67,68], as a classical network in semantic segmentation, has achieved great success in landslide detection tasks [34,38,39]. It has been demonstrated that the skip connection in U-Net effectively alleviates the detail loss problem due to convolution, and the residual structure prevents the vanishing gradient when mining deeper information [38].…”

“…Furthermore, Qi et al [38] proved that the U-Net with ResNet50 can improve the performance of the model on rainfallinduced landslide detection. Ghorbanzadeh et al [39] firstly used the free Sentinel-2 data in landslide identification by evaluating the performance of U-Net and ResU-Net in three different landslide areas, the result indicated that ResU-Net obtained the highest F1-score. To alleviate the lack of generalization of the model caused by various landslide morphologies, Yi et al [18] introduced an attention mechanism to assign weights for important feature maps in ResU-Net, the F1-score on the proposed model was improved by 7% compared to ResU-Net.…”

Mapping landslides through a temporal lens: An insight towards multi-temporal landslide mapping using the U-Net deep learning model

“…We extracted patches of 128 x 128 pixels from the input satellite images, as suitable input for CNNs, based on Ghorbanzadeh et al (2021) and Prakash et al (2021), who reported optimal accuracies of F1score, Precision, and Recall using this patch size. By rasterizing the manual inventory of the different years with 5 meter grid cells, the associated binary masks were created.…”

“…The U-Net model has been used extensively for landslide detection (Ronneberger, et al, 2015) due to its robust network structure and segmented pixels as outputs (Ghorbanzadeh, et al , 2021;Prakash et al , 2021;Zhang et al, 2018). The U-Net model (Figure 3) has many advantages.…”

A Novel Method for Extracting Time Series Information of Deformation Area of a Single Landslide Based on Improved U-Net Neural Network

“…But these methods are lacking in accuracy and recognition speed. Some scholars use u-net neural network to extract landslide deformation information (Ghorbanzadeh et al, 2021). Sanghoon Lee et al, used U-Net to quantitative spatial analysis on whole slide images (Lee et al, 2020).…”