Automatic extraction of building footprints from images is one of the vital means for obtaining building footprint data. However, due to the varied appearances, scales, and intricate structures of buildings, this task still remains challenging. Recently, the vision transformer (ViT) has exhibited significant promise in semantic segmentation, thanks to its efficient capability in obtaining long-range dependencies. This article employs the ViT for extracting building footprints. Yet, utilizing ViT often encounters limitations: extensive computational costs and insufficient preservation of local details in the process of extracting features. To address these challenges, a network based on an enhanced ViT (EViT) is proposed. In this network, one convolutional neural network (CNN)-based branch is introduced to extract comprehensive spatial details. Another branch, consisting of several multiscale enhanced ViT (EV) blocks, is developed to capture global dependencies. Subsequently, a multiscale and enhanced boundary feature extraction block is developed to fuse global dependencies and local details and perform boundary features enhancement, thereby yielding multiscale global-local contextual information with enhanced boundary feature. Specifically, we present a window-based cascaded multihead self-attention (W-CMSA) mechanism, characterized by linear complexity in relation to the window size, which not only reduces computational costs but also enhances attention diversity. The EViT has undergone comprehensive evaluation alongside other state-of-the-art (SOTA) approaches using three benchmark datasets. The findings illustrate that EViT exhibits promising performance in extracting building footprints and surpasses SOTA approaches. Specifically, it achieved 82.45%, 91.76%, and 77.14% IoU on the SpaceNet, WHU, and Massachusetts datasets, respectively. The implementation of EViT is available athttps://github.com/dh609/EViT.

Underground coal fires are global catastrophes that result in energy waste, carbon emission, and eco-environment pollution. Remote sensing (RS) detection is essential for underground coal fire extinguishing engineering, and the most used is thermal infrared (TIR) RS. It can well obtain the thermal anomalies of land surface temperature (LST), which is the most direct surface feature of underground coal fires. However, most studies using TIR RS simply delineate underground fire sources vertically according to LST anomalies, which has relatively little impact when initially determining coal fire area locations on the large scale. As for the precise location of small-scale subsurface fire sources, the deviation between subsurface fire source locations inferred and real locations could lead to errors or even mistakes to fire extinguishing engineering. There is a lack of subsurface fire source evolution model reconstruction method, and the spatiotemporal correlations characteristic of LST thermal anomalies and underground fire sources have not yet been discussed. To this end, taking Miquan coalfield (Western China) as an example, a 3-D empirical Bayesian Kriging (EBK3D) method is first proposed to reconstruct the 4-D temperature fields of underground fire sources. Then, the feasibility of the vertical correspondence approach to inferring small-scale subsurface fire sources through LST thermal anomalies detected by unmanned aerial vehicle TIR RS and satellite TIR RS is analyzed. Finally, the spatiotemporal correlation characteristic of LST thermal anomalies and subsurface fire sources is analyzed. As the results show, it is feasible to reconstruct the underground fire source evolution model by the EBK3D method. The reconstructed 4-D temperature fields can dynamically reflect the evolutionary states of underground fire sources in three time periods, with cross-validated root mean square errors of 52.2 °C, 49.6 °C, and 37.1 °C and$R^{2}$of linear regressions of 0.925, 0.9145, and 0.8429, respectively. The LST thermal anomalies show a significant spatiotemporal delay with respect to the subsurface fire source evolution. This makes the locations of the underground fire sources traced by the vertical correspondence method deviate from the real ones. The offsets of underground fire sources relative to surface thermal anomalies in the coal seam strike and dip directions for different time periods at depths of (T1: −44.43 m, T2: −27.72 m, and T3: −20.04 m) are (T1: 73.80 m, T2: 52.33 m, and T3: 45.06 m), and (T1: 16.79 m, T2: 17.27 m, and T3: 24.82 m), respectively.$R^{2}$’s for the linear regression model of the offset averages in three directions versus time and fire source size are (0.9247, 0.7949, and 0.9564) and (0.8739, 0.85 and 0.9152), respectively.

This letter presents a multiscale convolutional neural network with color vegetation indices (MCCNN) for semantic labeling of point cloud directly in a 3-D model. First, color vegetation indices are calculated for each point with RGB information. Second, based on classic Point convolutional neural network (CNN), a new multiscale network is designed to incorporate multiscale information through the spatial contexts of different sizes around each point by setting different convolution of kernels$K$, and then multiscale features produced by different convolutional layers are aggregated and unsampled. Finally, via Fully Connection layer and Softmax classifier, each point is labeled. Two different datasets, Semantic3D and Vaihingen3D, are used to evaluate the performance of the proposed method, and the results are compared with those produced by other existing approaches. Experimental results indicate that the proposed method achieves 84.5% in terms of overall accuracy on Semantic3D, and 85.2% on Vaihingen3D, which is the highest among the considered methods.

Spatiotemporal fusion technique provides a cost-efficient way to achieve dense time series observation. Among all categories of spatiotemporal fusion methods, the weight function-based method attracted considerable attention. However, this kind of method selects similar pixels in a regular window without considering the distribution of features, which will weaken its ability to preserve the structure information. Besides, the weight function-based method carries out pixel-by-pixel fusion computation, which leads to computational inefficiency. To solve the aforementioned issues, a flexible object-level (OL) processing strategy is proposed in this article. Three popular spatiotemporal fusion methods include the spatial and temporal adaptive reflectance fusion model (STARFM), the enhance STARFM (ESTARFM) and the three-step method (Fit-FC) were selected as examples to analyze and validate the effectiveness of the OL processing strategy. Four study sites with different surface landscapes and change patterns were adopted for experiments. Experimental results indicated that the OL fusion versions of STARFM, ESTARFM, and Fit-FC can better preserve the structural information, and were 102.89–113.71, 92.77–115.73, and 30.51–36.15 times faster than their original methods. Remarkably, the OL fusion versions of Fit-FC outperform all competing methods in one-pair case fusion experiments, especially in Poyang lake wetland (PY) area (root mean square error (RMSE) is 0.0343 versus 0.0380, correlation coefficient ($r$) is 0.7469 versus 0.6986 compare with Fit-FC). Additionally, the OL processing strategy can also be adopted to enhance other methods which use the principle of combining similar adjacent information. The program and test data are available athttps://github.com/Andy-cumt.

In this letter, we propose a change detection method based on Gabor wavelet features for very high resolution (VHR) remote sensing images. First, Gabor wavelet features are extracted from two temporal VHR images to obtain spatial and contextual information. Then, the Gabor-wavelet-based difference measure (GWDM) is designed to generate the difference image. In GWDM, a new local similarity measure is defined, in which the Markov random field neighborhood system is incorporated to obtain a local relationship, and the coefficient of variation method is applied to discriminate contributions from different features. Finally, the fuzzy c-means cluster algorithm is employed to obtain the final change map. Experiments employing QuickBird and SPOT5 images demonstrate the effectiveness of the proposed approach.

This paper presents a novel adaptive fuzzy local information c-means (ADFLICM) clustering approach for remotely sensed imagery classification by incorporating the local spatial and gray level information constraints. The ADFLICM approach can enhance the conventional fuzzy c-means algorithm by producing homogeneous segmentation and reducing the edge blurring artifact simultaneously. The major contribution of ADFLICM is use of the new fuzzy local similarity measure based on pixel spatial attraction model, which adaptively determines the weighting factors for neighboring pixel effects without any experimentally set parameters. The weighting factor for each neighborhood is fully adaptive to the image content, and the balance between insensitiveness to noise and reduction of edge blurring artifact to preserve image details is automatically achieved by using the new fuzzy local similarity measure. Four different types of images were used in the experiments to examine the performance of ADFLICM. The experimental results indicate that ADFLICM produces greater accuracy than the other four methods and hence provides an effective clustering algorithm for classification of remotely sensed imagery.

The level set method, because of its implicit handling of topological changes and low sensitivity to noise, is one of the most effective unsupervised change detection techniques for remotely sensed images. In this letter, an expectation-maximization-based level set method (EMLS) is proposed to detect changes. First, the distribution of the difference image generated from multitemporal images is supposed to satisfy Gaussian mixture model, and expectation-maximization (EM) is then used to estimate the mean values of changed and unchanged pixels in the difference image. Second, two new energy terms, based on the estimated means, are defined and added into the level set method to detect those changes without initial contours and improve final accuracy. Finally, the improved level set method is implemented to partition pixels into changed and unchanged pixels. Landsat and QuickBird images were tested, and experimental results confirm the EMLS effectiveness when compared to state-of-the-art unsupervised change detection methods.

This letter presents a semi-automatic approach to delineating road networks from very high resolution satellite images. The proposed method consists of three main steps. First, the geodesic method is used to extract the initial road segments that link the road seed points prescribed in advance by users. Next, a road probability map is produced based on these coarse road segments and a further direct thresholding operation separates the image into two classes of surfaces: the road and nonroad classes. Using the road class image, a kernel density estimation map is generated, upon which the geodesic method is used once again to link the foregoing road seed points. Experiments demonstrate that this proposed method can extract smooth correct road centerlines.

This letter presents a two-step method for urban main road extraction from high-resolution remotely sensed imagery by integrating spectral-spatial classification and shape features. In the first step, spectral-spatial classification segments the imagery into two classes, i.e., the road class and the nonroad class, using path openings and closings. The local homogeneity of the gray values obtained by local Geary's C is then fused with the road class. In the second step, the road class is refined by using shape features. The experimental results indicated that the proposed method was able to achieve a comparatively good performance in urban main road extraction.

The soft-then-hard subpixel mapping (STHSPM) algorithm is a type of subpixel mapping (SPM) algorithm consisting of soft class value (between 0 and 1) estimation and hard class allocation for each subpixel. This letter presents a new class allocation method for STHSPM algorithm. As an extension of our previous work in which subpixels for classes are decided in units of classes (UOC), the new approach, named adaptive UOC (AUOC), improves UOC with adaptive visiting order of classes. In AUOC, the visiting order of classes within each coarse pixel is determined based on the local structure rather than the global structure in UOC. Experiments on three remote sensing images show that AUOC is able to improve UOC in terms of SPM accuracy, particularly for SPM with small zoom factors.

This letter presents a novel spatial-attraction-based Markov random field (MRF) (SAMRF) approach for high spatial resolution multispectral imagery (HSRMI) classification. First, the initial class label and class membership for each pixel are obtained by applying the maximum likelihood classifier (MLC) classification for the HSRMI. Second, to reduce the oversmooth classification in the traditional MRF, an adaptive weight MRF model is introduced by integrating the spatial attraction model into the traditional MRF. Finally, the initial classification map, generated in the first step, will be refined though the SAMRF regularization. Two different experiments were performed to evaluate the performance of the SAMRF, in comparison with standard MLC and MRF. Experimental results indicate that the SAMRF method achieved the highest accuracy, hence, providing an effective spectral-spatial classification method for the HSRMI.

This letter presents a novel spatial features extraction method for the high spatial resolution multispectral imagery (HSRMI) classification. First, Canny filter algorithm is applied to extract the edge information to obtain the fuzzy edge map. Secondly, adaptive threshold value for each pixel's homogeneous region (PHR) calculation is determined based on the fuzzy edge map and original image. Next, the PHR for every pixel is obtained based on the fuzzy edge map, adaptive threshold value and original image. And then, the pixel shape feature set (PSFS) is extracted based on the PHR. Lastly, SVM classifier is applied to classify the hybrid spectral and PSFS. Two different experiments were performed to evaluate the performance of PSFS, in comparison with spectral, gray level co-occurrence matrix (GLCM) and the existing pixel shape index (PSI). Experimental results indicate that the PSFS achieved the highest accuracy, hence, providing an effective spectral-spatial classification method for the HSRMI.

Road centerline extraction from remotely sensed imagery can be used to update a Geographic Information System (GIS) database. The common road extraction from high-resolution imagery is based on spectral information only; it is difficult to separate road features from background completely, and a thinning algorithm always results in short spurs which reduce the smoothness of the road centerline. To overcome the aforementioned shortcomings of the common existing road centerline algorithms, this letter presents a new method to extract the road centerline from high-resolution imagery based on shape features and multivariate adaptive regression splines (MARS), in which potential road segments were obtained based on shape features and spectral feature, followed by MARS to extract road centerlines. Two experiments are performed to evaluate the accuracy of the proposed method.

This paper presents a novel fuzzy-topology-integrated support vector machine (SVM) (FTSVM) classification method for remotely sensed images based on the standard SVM. Induced threshold fuzzy topology is integrated into the standard SVM. First, the optimal intercorrelation coefficient threshold value is applied to decompose an image class in spectral space into the three parts: interior, boundary, and exterior in fuzzy-topology space. The interior-class pixels are then classified as predefined classes based on maximum likelihood. The exterior-class pixels are ignored. The fuzzy-boundary-class pixels which contain misclassified pixels are reclassified based on the fuzzy-topology connectivity theory. As a result, misclassified pixel problems, to a certain extent, are solved. Two different experiments were performed to evaluate the performance of the FTSVM method, in comparison with standard SVM, maximum likelihood classifier (MLC), and fuzzy-topology-integrated MLC. Experimental results indicate that the FTSVM method performs better than the standard SVM and other methods in terms of classification accuracy, hence providing an effective classification method for remotely sensed images.