Landslides are one of the most frequent and destructive geological disasters, often causing significant threats to human life and infrastructure. Landslide susceptibility assessment (LSA) plays a vital role in disaster prevention and risk management. Graph convolutional networks (GCNs) have demonstrated strong potential in LSA due to their ability to model graph-structured data. However, existing GCN-based models often suffer from feature aggregation issues, where the neighborhood aggregation of landslide conditioning factors (LCFs) leads to information loss and reduced predictive performance. To address this limitation, we propose an internal and external decoupled GCN (IED-GCN) based on superpixel segmentation. First, the simple linear iterative clustering (SLIC) algorithm is employed to segment the study area into superpixels, creating more meaningful spatial units. Then, the proposed model applies internal graph convolution (IGC) to capture local feature dependencies and external graph convolution (EGC) to model global feature relationships. Furthermore, edge decoupling is incorporated into the EGC to mitigate overfitting and enhance the model’s generalization capability. The effectiveness of IDE-GCN is validated using a dataset comprising historical landslide occurrences and 13 LCFs in the Three Gorges Reservoir area. The dataset is divided into 70% of the data for training and 30% for testing. Experimental results demonstrate that our IED-GCN significantly outperforms conventional models such as convolutional neural networks (CNN), residual convolutional networks (ResNet), dense convolutional networks (DenseNet), and miniGCN, achieving F1 score, Kappa coefficient, and AUC values of 0.9839, 0.9675, 0.9838. In summary, our IED-GCN effectively addresses the challenge of feature aggregation in LSA tasks, offering superior performance in geological disaster prediction with complex spatial data.

The “black-box” nature of machine learning (ML) and deep learning (DL) models has raised concerns about the trustworthiness of landslide susceptibility mapping (LSM) results among users. Existing studies have applied many techniques to interpret LSM models, but they predominantly focused on individual model interpretations, lacked comparisons of interpretation results across different models, and failed to fully explore the potential of explainable artificial intelligence (AI) techniques in LSM. This study develops an innovative model interpretation framework based on the Shapley additive explanation (SHAP) method and different ML and DL models, to analyze the decision mechanisms differences and discuss the geospatial heterogeneity of landslide conditioning factors (LCFs). A geospatial database is constructed, including historical landslides, 16 common LCFs, and three earthquake-related LCFs for two study areas: Zigui and Jiuzhaigou. The data are then divided into training and testing sets in a 7:3 ratio for four models: random forest (RF), extreme gradient boosting decision tree (XGBoost), residual network, and densely connected convolutional networks (DenseNets). Finally, global and local interpretations are provided using the SHAP method. The analysis indicates that: 1) XGBoost consistently outperforms the other models in both study areas, achieving Kappa coefficient (Kappa), overall accuracy (OA), and area under the receiver operating characteristic curve (AUC) values of 0.9416, 0.9738, and 0.9757 for Zigui, and 0.8525, 0.9337, and 0.9312 for Jiuzhaigou and 2) the global interpretation shows that the same LCFs play different roles in the XGBoost and DenseNet, reflecting different decision mechanisms among LSM models. Moreover, local interpretations demonstrate that the same LCFs contribute differently in the two areas, highlighting the geospatial heterogeneity in LSM.

Recognizing landslides effectively is crucial for disaster prevention and post-disaster rescue operations. Our SwinLiteKD, an innovative knowledge distillation network based on Swin-Transformer, addresses challenges related to model runtimes and inefficiencies in current deep learning approaches. Validated in a landslide-prone region in Zigui County, Hubei Province, China, our proposed method incorporates landslide influencing factors to significantly enhance model performance. Compared to ResNet50, Swin-Transformer, and DeiT, SwinLiteKD achieves superior Overall Accuracy (OA: 97.0000%), Precision (97.1698%), Recall (96.9999%), F1 (97.0848%), and Kappa (93.99%). With the lowest number of FLOPs, our model ensures crucial computational efficiency in landslide recognition after geological disasters, requiring only 4.4987 GFLOPs. This is 0.1553 GFLOPs, 0.0723 GFLOPs, and 0.3597 GFLOPs less than ResNet, Swin, and DeiT, respectively. SwinLiteKD demonstrates excellent adaptability in scenarios demanding swift landslide recognition post-geological disasters.

Landslide disasters are highly destructive and harmful, seriously threatening the safety of human life and property. Timely and accurate identification is crucial for disaster reduction and relief. Currently, Convolutional Neural Network (CNN) has good ability to extract local features of images, and Transformer can capture long-distance dependencies when processing sequence data. However, in the method that combines CNN and Transformer, the feature information of both cannot be fully utilized. Therefore, this article proposes an improved MLTransUNet method, which combines the information obtained by the convolution and Transformer modules through multi-level feature fusion. This method was applied to the Zigui to Badong section of the Three Gorges Reservoir area in Hubei Province. Experimental results showed that MLTransUNet achieved F1 score, Kappa and MIoU of 0.8560, 0.8499, and 0.8680, respectively. Compared with the TransUNet model, it improved by 2.62%, 2.75%, and 2.08%, respectively. This indicates that it has better performance in landslide recognition and provides technical support for landslide disaster identification and management.

Prompt and accurate identification of landslides immediately after a disaster is crucial for evaluating the subsequent hazards and risks in the affected areas. The proposed method in this paper aims to detect landslides and accurately determine their extent using semantic segmentation. Many existing models rely on multi-temporal or geological data to enhance their accuracy. However, the inclusion of a large amount of data introduces additional parameters, consumes significant computational resources, and requires lengthy training time. To address these challenges, this paper introduces a lightweight network called BiSeNet for landslide recognition. By incorporating modifications to the network’s skip connections, dense connections are added to BiSeNet. This modification aims to improve the network’s performance in landslide detection while minimizing computational requirements and training time. Experimental results revealed that the BiSeNet-DenseNet model achieved an F1-score value of 0.8856, showing a significant 30.35% improvement over BiSeNet-Xception.

Remote sensing building change detection (RSBCD) detects changes in the spatial distribution of buildings which is of great significance for urban planning and construction. Existing deep learning-based RSBCD methods usually suffer from low object completeness and erroneous detection problem, mainly due to insufficient utilization of difference information between bi-temporal images. To address the above issues, this article proposed a new Siamese ensemble difference attention network (SEDANet) for RSBCD tasks in very-high resolution (VHR) images. Firstly, the key module ensemble difference attention module (EDAM) is designed to effectively extract difference representation between the bi-temporal features and filter out irrelevant changes. EDAM calculates difference map of bi-temporal features and transforms the extracted change information into trainable difference attention weights. The output weights from EDAM works as a guidance for both spatial and channel visual attention process, which enables the network to focus on foreground building changes and further resolve erroneous attention problems in existing RSBCD methods. The Siamese structure is adopted to better represent bi-temporal features, and convolutional blocks are replaced with residual convolution blocks (RCBs) to speed up network fitting and prevent gradient explosion or descent. We conduct comprehensive experiments on three benchmark datasets. Both visual and quantitative results show that our proposed SEDANet is superior to other eight state-of-the-art networks. Especially on GZ-CD dataset, SEDANet outperforms other comparison methods by 3%-8%. In addition, the effectiveness of EDAM module is also discussed through a series of ablation studies.

A novel InSAR-based time-series prediction framework was proposed to analyze ground subsidence in underground coal mines in Yuzhou City. The study utilized Sentinel-1A descending track data and employed a deep learning model. Initially, the SBAS-InSAR technique was employed to examine the spatiotemporal evolution patterns of ground subsidence in the designated area. The wavelet transform algorithm was applied to extract the trend term displacement and periodic term displacement. Subsequently, a univariate LSTM network was utilized to predict the trend term displacement, while a multivariate CNN-LSTM network was employed to predict the periodic displacement. The preliminary results demonstrated a high level of agreement between the time-series predictions obtained through this framework and the InSAR monitoring values.

Due to its unique geographical location and development potential, the Yangtze River Economic Belt (YEB) has emerged as one of the most economically developed regions in China. The study employs remote sensing data to conduct a coupled coordination analysis of the ecological environment and urbanization processes of the three major urban agglomerations in the YEB using the Coupled Coordination Model (CCD). The results show that (1) the mean RSEI values of the three urban agglomerations initially declined, then fluctuated upwards; (2) the CNLI values of all three urban agglomerations maintained a continuous increase, with the urbanization level of the Yangtze River Delta urban agglomeration being significantly higher than that of the other two urban agglomerations; and (3) The degree of coupling and coordination of the Yangtze River Delta urban agglomeration was significantly higher than that of the Chengdu-Chongqing urban agglomeration and the middle reaches of the Yangtze River urban agglomeration.

Surface deformation is a complex geological phenomenon with potential threats to urban construction and human safety. Time-series prediction of surface deformation has an important role in mitigating the impact of such hazards. This study utilizes the small baseline subset interferometric synthetic aperture radar method (LiCSBAS) to monitor the long-time surface deformation in the main urban area of Kunming, and constructs a deep neural network model (TCN-GRU) to predict typical surface deformation points in the study area. The results show that Kunming city experienced a surface deformation rate of -44.23-17.65 mm/y from March 2018 to July 2022, along with the presence of five subsidence funnels. The TCN-GRU model demonstrates the best short-term prediction performance on different datasets, and significantly outperforms other traditional deep learning models. The results of this paper can serve as a valuable reference for the study of urban surface deformation.

Natural hazards frequently threaten human life, ecosystem, and economy all around the world, while landslides are the most destructive ones. Landslide susceptibility mapping (LSM) is an effective strategy to determine the probability of future landslide events by involving a comprehensive analysis of various factors, including geological environmental, historical landslides and the landslide physical laws. As artificial intelligence techniques are becoming more popular in LSM, it is important to understand how decisions are made by these models. This study aims to use representative ML (ML) and deep learning (DL) models (include random forest (RF), support vector machine (SVM), residual neural networks (ResNet) and Densely connected convolutional networks (DenseNet)) to map the landslide susceptibility of the study area in Zigui, and then visualize the decision-making process of the model through an explainable artificial intelligence (XAI) technology, so as to provide more transparency and reliability about the occurrence of landslides.

Remote sensing change detection (RSCD) can recognize large-scale spatial building distribution changes, saving a lot of manpower and material resources compared to field surveys or manual visual interpretation. The existing deep learning based change detection (CD) networks are commonly improved from segmentation networks. Although the overall accuracy of CD tasks has been greatly improved, there are still noticeable shortcomings in terms of the object completeness and edge details. For this issue, we proposed a Siamese Residual Network (SRNet) based on U-Net for VHR images CD tasks. We first constructed a Siamese network for better utilization of bi-temporal images information. And then we added residual connection to form a convolutional block for faster model convergence and easier training difficulties. We conduct comprehensive experiments on two benchmark datasets. Both the visual and quantitative results show that our proposed SRNet is superior to the other five state-of-the-art networks.

High-quality landslide susceptibility mapping (LSM) is an important step in landslide disaster prevention and mitigation. In order to explore and evaluate the practical performance of a convolutional and transformer hybrid model in landslide susceptibility mapping, this study selected 202 historical landslides and seven conditioning factors to construct a geographical spatial dataset for LSM. The hybrid model called the compact convolutional transformer (CCT) model was used to perform LSM in the Three Gorges Reservoir area in China, and it was compared with vision transformer (ViT), convolutional neural network (CNN), and residual network (ResNet). The results show that the CCT model has the best overall performance, achieving the highest accuracy in multiple statistical metrics, and improving the accuracy of LSM. This study provides a new approach for obtaining high-quality LSM using a convolutional and transformer hybrid model.

The acquisition of information on land occupation in mining areas is important for ecological management and restoration of mining areas. The current work has the problem of low extraction accuracy due to the small number of typical land category samples and simple network structure. In order to improve the accuracy of extraction of land use information in mining areas, this paper uses the data from the Gaofen-2 satellite, combines the dense connected neural network (DenseNet) with object-oriented ideas to extract mining areas in Yuzhou City, Henan Province. The results are compared with convolutional neural network (CNN) and support vector machine (SVM). The results show that the overall accuracy and kappa coefficient of the proposed DenseNet is 87.46% and 0.84, which are 0.06%, 2.63% and 0.004, 0.039 higher than that of the CNN and SVM respectively. The proposed DenseNet performs the best in the extraction of open-pit and waste-dump area, which indicating that it can provide promising advantages in mine information extraction, and can provide technical support for environmental monitoring and scientific management of mining areas.

Sparse unmixing (SU) has been widely applied to remotely sensed hyperspectral images interpretation. Compared with traditional unmixing algorithms, SU does not need to extract pure signatures (endmembers) from the image. The endmember matrix is constructed by directly selecting spectra from a known library which is used to estimate the fractional abundances associated with endmembers. This avoids the problem of extracting virtual endmembers without physical meaning. However, SU does not generally include spatial information, which may limit its performance. In order to address this limitation and include local spatial information, low-rank and sparse features in local regions can be exploited. In this paper, we include spatial information in the traditional SU algorithm by extracting low rank and spatial information based on superpixels, and further propose an algorithm named superpixel-based collaborative sparse and low-rank regularization for sparse unmixing (SCLRSU) to improve the performance of the traditional spatial regularization-based SU methods. In our proposed method, we combine superpixel segmentation and structural sparsity. Experiments are carried out on two simulated datasets and two real hyperspectral image datasets, and our results are compared with those obtained by traditional SU methods. Our results indicate that our newly proposed method provides very competitive performance.

Identification of lineament structure plays a vital role in determining the metallogenic area and distribution of the geologic structure. Edge detection methods are mostly used to recognize the lineaments and define the geologic boundaries. Cooperatively using edge detection results of the gravity, magnetic and remote sensing data to recognize lineaments would obtain more geologic information. In this paper, new edge detectors of potential field derivatives are proposed to determine the sources’ boundary, named second tilt derivative, tilt of vertical derivative, and normalized second vertical derivative, respectively. Presented approaches are characterized by producing zero amplitude over sources’ edges and equalizing anomalies from different depths. Compared with original edge detection techniques including other second derivative methods, synthetic examples reveal significant superiorities of suggested approaches in providing more accurate and sharper edges and are especially effective in distinguishing superimposed anomalies. The experiments also demonstrate that the normalization to the edge detectors will make images cleaner and geologic edges more easily captured. Applied to airborne gravimetric and magnetic data in the Pobei area (NW China), the proposed methods display more geologic details and lineaments. Canny, Sobel, and Prewitt operators are applied to extract boundaries of remote sensing image. Lineaments picked by the three different types of data are combined collectively to get a comprehensive lineaments structure interpretation.

Landslide is a widely developed disaster in the world, which brings serious harm to economy and people. How to detection landslide intelligently, quickly and accurately has become the focus and difficulty of researchers in the field of geological hazards. In recent years, with the development of remote sensing technology and computer technology, researchers have built various landslide detection automatic or semi-automatic models. Compared with traditional interpretation methods and machine learning methods, various landslide detection models based on deep learning algorithm such as convolutional neural network (CNN) perform better and more intelligent. DenseN et is an improved CNN algorithm based on image classification, which was proposed in 2017. In this paper, we transferred the DenseNet and CNN to construct landslide detection models. After accuracy comparison, it can be shown that DenseNet performs better than CNN. The Kappa coefficient and F1 score is of 0.965, 0.995 for DenseNet, and 0.908, 0.896 for CNN, respectively. The results showed that the DenseN et model has a higher landslide detection accuracy and generalization ability.

Earthquake-induced landslide is the most common geological disaster caused by earthquakes, which seriously threatens the safety of human life and property. Rapid access to landslide information after the earthquake is the key to disaster mitigation and relief. The existing landslide extraction methods all require a lot of manual intervention and cannot provide timely and effective earthquake rescue information. In order to solve the above problems, this paper improves the traditional U -Net algorithm by feature expansion and proposes a FE-U-Net. This new method is applied to the landslide in Jiuzhaigou County, Sichuan Province, China by combining multi-source remote sensing images. The experimental results show that the F1-score, Kappa coefficient and mIOU values obtained by the proposed FE-U-Net are 95.02%, 94.23% and 94.47%, respectively, which are all about 3% higher than the traditional U -Net. In addition, the model training time is less than 10s longer than the traditional U -Net model.

Hyperspectral remote sensing plays role in the field of earth observation research because of rich spatial, radiation and spectral information. With the rapid development of deep learning, deep neural networks are widely used in hyperspectral remote sensing image classification tasks, but at the same time, a series of difficulties have arisen, such as high demand for training samples, time-consuming model training. Convolutional neural network (CNN) is well known for its capability of feature learning and has demonstrated excellent performance in hyperspectral image classification. In this paper, the one-dimension CNN (1D-CNN) modules was added after the two-dimension CNN (2D-CNN), so the complete network structure contains two different dimension CNN, called multi-dimension CNN (MD-CNN). The proposed 1D-CNN blocks can continue to learn contextual features and is expected to have more discriminative power. Experimental results with hyperspectral image benchmark datasets demonstrate that the proposed method can outperform the state-of-the-art CNN-based classification methods.

The sparse representation has been introduced for hyperspectral anomaly detection methods. However, the window parameter tuning and anomaly contamination problems are still the main issues with the background dictionary. In order to solve these problems, this paper proposed the joint sparse representation and multi-task learning method (JSM) for anomaly detection. This method utilizes a global background dictionary construction method to avoid the above window parameter tuning and anomaly contamination problems. Besides, the multi-task learning technology is employed to explore the hyperspectral images similarity within adjacent single-band images. Experiments were carried out on two hyperspectral images, and it was founded that JSM method shows a better detection performance than the other anomaly detection methods.

With the rapid development of deep learning in recent years, the field of remote sensing image processing has gradually started to use some deep learning algorithms to achieve intelligent and fast processing of images, and the results have improved to a certain extent compared to traditional methods. The U-Net convolutional neural network was proposed used in medical image segmentation in 2015. Based on the previous work, we transferred the U-Net to remote sensing image segmentation to realize the pixel level semantic segmentation of remote sensing image end-to-end. Through U-Net training and learning on GF-2 remote sensing image, the overall accuracy of training sets is 93.83%, while the overall accuracy of the test data is 82.27%, the kappa coefficient is 0.7721, and the Mean intersection Over Union (MiOU) is 0.6405. The results showed that the experiment has high segmentation accuracy and generalization ability.

In this paper, we propose a novel approach based on a symmetric fully convolutional network within pyramid pooling (FCN-PP) for landslide mapping (LM). The proposed approach has three advantages. Firstly, this approach is automatic and insensitive to noise because multivariate morphological reconstruction (MMR) is used for image preprocessing. Secondly, it is able to take into account features from multiple convolutional layers and explore efficiently the context of images, which leads to a good tradeoff between wider receptive field and the use of context. Finally, the selected pyramid pooling module addresses the drawback of single-scale pooling employed by convolutional neural network (CNN), fully convolutional network (FCN), U-Net, etc. Experimental results show that the proposed FCN-PP is effective for LM, and it outperforms state-of-the-art approaches in terms of four metrics, Precision, Recall, F -score, and Accuracy.

Mineral resources are an important material basis for economic and social development. The development and utilization of mineral resources is also an inevitable requirement for modernization construction. However, it is always inevitable to have a negative impact on the natural environment in the process of mineral extraction. The phenomenon of landscape destruction, including open pit, scrap slag heap, tailings reservoir and so on, is common. Especially the damage to the environment caused by open pit is particularly serious. In order to achieve fast and accurate detection of open pits, this paper uses the method of combining convolution neural network with object-oriented thought to extract the open pit in the mining area of Yuzhou, Henan province. Accuracy evaluation of classification results based on actual field survey data and land use data. The final total accuracy is 91.18%, kappa coefficient is 0.89, which shows the usability and advantages of this method in the application of open pit extraction in mining area.

Mine environmental problems are intensifying, and the environmental monitoring and evaluation of the mining area is an indispensable part of mine management. Previous researches have obtained effective results by evaluating the mining ecological environment through the Remote Sensing Ecological Index (RSEI). However, the calculation does not take into account that the environmental influence is regional under natural conditions of RSEI, and the geographical location of the mining area has different effects on its surrounding ecological environment. Aiming at the complex research area of large-scale object types, this paper proposes an improved RSEI named Moving Window-based Remote Sensing Ecological Index (MW-RSEI) which based on RSEI factor and moving window evaluation unit. The results show that MW-RSEI is consistent with RSEI, and it can display more environmental information of mines, which is more in line with natural laws and provides an effective basis for environmental assessment of mining areas.

Using multiple endmember spectral mixture analysis, six remote sensing images of Landsat data in Wuhan area through 1991 to 2013 are unmixed to investigate the urbanization process. The result indicates that urbanization process has resulted in the transformation of the natural landscape to an anthropogenic urban built-up area. However, this transformation has not been smooth. In addition, the urbanization pattern changed in time and space. From 1991 to 2000 and 2009 to 2013, the urban growth pattern consisted of new land development, whereas from 2000 to 2005, it was focused more on urban redevelopment. Spatially, the expansion of the built-up area occurred outward from the geographic center of Wuhan. The process in Hankou was dependent on the structure of the old town and developed layer by layer in an arc. The built-up area grew at a slow rate throughout Hanyang, while in Wuchang, it expanded to the east as a result of the economic developing area. The correlation between the total built-up area with social statistic proves that the urbanization process leads to the population growth and economic explosion.