Landslides are among the most common natural disasters globally, posing significant threats to human society. In recent years, deep learning (DL) has been widely applied to rapid landslide detection tasks. However, large-scale, multi-area, and multi-sensor landslide datasets for DL landslide detection are still relatively scarce. Most existing datasets adopt a fixed patch size, overlooking the variations in spatial resolution and landslide scale in remote sensing images, thereby limiting the performance of DL models. To address these limitations, we construct a Large-scale Multi-source High-resolution Landslide Dataset (LMHLD). LMHLD collects remote sensing images from five different satellite sensors, covering seven study areas around the world. LMHLD comprises 25,365 image patches of varying sizes and includes 32,296 annotated landslide instances across diverse geographical environments. Additionally, we propose a Semi-Adaptive Patch Size Selection method (SAPSS), which adaptively selects optimal patch sizes for different study areas. Furthermore, we design a training module, LMHLDpart, which enables the seamless integration of multiple heterogeneous sub-datasets within LMHLD, thereby enhancing the flexibility and robustness of DL models trained on LMHLD. Finally, we demonstrated in four evaluation experiments that LMHLD has the potential to become a benchmark dataset for landslide detection. LMHLD provides a strong foundation for DL models, accelerates the development of DL in landslide detection, and serves as a valuable resource for landslide prevention and mitigation efforts. LMHLD is open access and can be accessed through the link: https://doi.org/10.5281/zenodo.11424987.

Geological environment remote sensing (GERS) interpretation contributes to lithological mapping, disaster prediction, soil erosion monitoring, and so on. However, the rich diversity, complex distribution, interclass similarities, and uncertainties in data quality of geological elements pose challenges to GERS interpretation. In addition, current automatic feature extraction of GERS elements, which rely on deep learning (DL) and remote sensing (RS) information process technologies, often require sufficient labeled data. Due to the enormous labor cost and specialized expertise needed, labeled GERS samples are limited to training the data-driven models. To tackle the above challenges, we introduce the semi-supervised dual-model progressive self-training (DM-ProST) framework. This framework employs two DL networks with different initializations as evaluator models to correct each other. A sample filtering strategy is then implemented to evaluate the quality of unlabeled samples, selecting high-quality and reliable ones to expand the training set. In addition, a fully connected conditional random field (CRF) module is incorporated to optimize DL network prediction maps, thereby enhancing the boundary performance of segmentation results. The framework utilizes a multitask loss function that combines consistency loss with cross-entropy, enabling the models to learn discriminative GERS features. This process accurately generates pseudo-labels and achieves precise lithology mapping of GERS with a small amount of annotation samples. Finally, we conducted an experimental evaluation on the Landsat 8 dataset in Xinjiang, China, and massive experiments proved the effectiveness of DM-ProST.

Convolutional neural networks (CNNs) are widely used in hyperspectral image classification (HSIC) due to their exceptional performance. However, current methods for multiscale feature extraction typically rely on single-branch CNNs, potentially causing interference among features of varying scales. To mitigate this issue, we present a multibranch adaptive feature fusion network (MBAFFN) classification method. MBAFFN enhances feature uniqueness and improves the accuracy and reliability of classification results by extracting information at multiple scales through three parallel branches. Furthermore, to address the challenge of capturing global features within CNNs, we introduce a global detail attention (GDA) mechanism aimed at bolstering the network’s capability to capture comprehensive information. In addition, we mitigate the issue of neglecting center-pixel importance in convolution operations through a distance suppression attention (DSA) design. To effectively integrate outcomes from multiple branches, we propose a pixel-based adaptive feature fusion strategy, thereby increasing the proportion of features conducive to improved classification results. Lastly, auxiliary loss functions are employed to train the multibranch network. Experimental results on four benchmark datasets demonstrate the superiority of our approach over several state-of-the-art methods, particularly in managing imbalanced small samples. Furthermore, ablation studies validate the effectiveness of the proposed modules.

Multitemporal landslide inventory mapping plays a vital role in postdisaster reconstruction, landslide prevention, and regional ecosystem restoration. While deep learning methods have achieved great success in landslide detection tasks, previous landslide detection approaches hardly use unlabeled samples to optimize models to distinguish landslide changes in multitemporal applications due to insufficient labeled data across different periods. To address this issue, we propose a novel method called progressive label upgradation and cross-temporal style adaption (PluTsa) for unsupervised multitemporal landslide detection. At the interdomain level, we introduce a paired image-to-image cross-temporal domain style adaption strategy to reduce visual differences among multitemporal remote sensing images. Besides, a temporal-aware pairing constraint (tpc) strategy is designed to further mitigate uneven feature distribution problems and align domain features. At the intradomain level, we propose a novel progressive label upgradation (PLU) scheme to produce high-quality pseudolabels that guide the deep learning model to extract valuable landslide features by connecting the geographic locations of cross-temporal images. The proposed method is evaluated on two datasets, and extensive experimental results demonstrate that PluTsa significantly outperforms other state-of-the-art methods, indicating it has promising prospects in unsupervised landslide detection from multitemporal high-resolution images.

Due to the complex and highly heterogeneous land cover in urban areas, the single-temporal pixel-wise and parcel-wise classification cannot realize high-precision recognition of ground objects. Semantic segmentation of satellite image time series (SITS), can distinguish objects with similar spectral reflection and temporal evolution. But optical SITS have problems of uneven time-frequency distribution and incomplete, which makes it impossible to directly use existing models to carry out time series semantic segmentation. This study proposes a semantic segmentation network that combines optical and radar SITS, named Multi-Source Temporal Attention Fusion-Based Temporal-Spatial Transformer (MTAF-TST), to achieve high-precision land cover classification in urban areas. Firstly, MTAF-TST uses the Transformer spatial semantic segmentation module to extract the spatial context information of ground objects to realize pixel-level land cover classification, which relieves the salt-and-pepper phenomenon that is easy to occur in traditional pixel-by-pixel classification in complex scenes. Secondly, MTAF-TST uses the Transformer time feature extraction module to mine long-range time-dependent and high-level semantic information, overcoming the drawbacks of traditional convolutional and recurrent neural networks that cannot mine long-range time-dependent features of SITS. Finally, MTAF-TST uses a multi-source temporal attention fusion module to fuse the depth features of optical and radar SITS, which overcomes the shortcomings of traditional direct feature stitching methods that cannot make full use of time-correlated features, achieving high-precision land cover classification. The experimental results show that the MTAF-TST can realize the complementarity of radar and optical SITS in terms of timing integrity, color, texture, etc., and effectively improve the accuracy of SITS classification.

This study introduces four heterogeneous ensemble-learning techniques, that is, stacking, blending, simple averaging, and weighted averaging, to predict landslide susceptibility in Yanshan County, China. These techniques combine several state-of-the-art classifiers of convolutional neural network, recurrent neural network, support vector machine, and logistic regression in specific ways to produce reliable results and avoid problems with the model selection. The study consists of three main steps. The first step establishes a spatial database consisting of 16 landslide conditioning factors and 380 historical landslide locations. The second step randomly selects training (70% of the total) and test (30%) datasets out of grid cells corresponding to landslide and non-slide locations in the study area. The final step constructs the proposed heterogeneous ensemble-learning methods for landslide susceptibility mapping. The proposed ensemble-learning methods show higher prediction accuracy than the individual classifiers mentioned above based on statistical measures. The blending ensemble-learning method achieves the highest overall accuracy of 80.70% compared to the other ensemble-learning methods.

Computing frameworks have been widely deployed to support global-scale services. A job typically has multiple sequential stages, where each stage is further divided into multiple parallel tasks. We call the set of all the tasks in a stage of a job a cotask. In this paper, we aim to minimize the average Cotask Completion Time (CCT) in cotask scheduling. To the best of our knowledge, there is no prior work on cotask scheduling for cloud computing. We propose the Cotask Scheduling Scheme (CSS), and take MapReduce as a representative of computing frameworks. CSS schedules cotasks following the Minimum Completion Time First (MCTF) policy, and we prove this problem is NP-hard. We formulate the model using the Integer Linear Programming (ILP), and solve it through an efficient heuristics based on ILP relaxation. Through real trace based simulations, we show that CSS is able to reduce the average CCT by up to 62.20% and 69.93% with traces from our testbed and from a large production cluster respectively.

In order to improve signal-to-noise ratio (SNR) and contrast-to-noise ratio, this paper introduces a local variance-controlled forward-and-backward (LVCFAB) diffusion algorithm for edge enhancement and noise reduction. In our algorithm, an alternative FAB diffusion algorithm is proposed. The results for the alternative FAB algorithm show better algorithm behavior than other existing diffusion FAB approaches. Furthermore, two distinct discontinuity measures and the alternative FAB diffusion are incorporated into a LVCFAB diffusion algorithm, where the joint use of the two measures leads to a complementary effect for preserving edge features in digital images. This LVC mechanism adaptively modifies the degree of diffusion at any image location and is dependent on both local gradient and inhomogeneity. Qualitative experiments, based on general digital images and magnetic resonance images, show significant improvements when the LVCFAB diffusion algorithm is used versus the existing anisotropic diffusion and the previous FAB diffusion algorithms for enhancing edge features and improving image contrast. Quantitative analyses, based on peak SNR, confirm the superiority of the proposed LVCFAB diffusion algorithm.