Deep learning has shown strong potential in modeling complex spatiotemporal dynamics. However, most existing methods depend on densely and uniformly sampled data, which is often unavailable in practice due to sensor and cost limitations. In many real-world settings, such as mobile sensing and physical experiments, data are burst-sampled with short high-frequency segments followed by long gaps, making it difficult to learn accurate dynamics from sparse observations. To address this issue, we propose Physics-Informed Multi-Scale Recurrent Learning (PIMRL), a novel framework specifically designed for burst-sampled spatiotemporal data. PIMRL combines macro-scale latent dynamics inference with micro-scale adaptive refinement guided by incomplete prior information from partial differential equations (PDEs). It further introduces a temporal message-passing mechanism to effectively propagate information across burst intervals. This multi-scale architecture enables PIMRL to model complex systems accurately even under severe data scarcity. We evaluate our approach on five benchmark datasets involving 1D to 3D multi-scale PDEs. The results show that PIMRL consistently outperforms state-of-the-art baselines, achieving substantial improvements and reducing errors by up to 80\% in the most challenging settings, which demonstrates the clear advantage of our model. Our work demonstrates the effectiveness of physics-informed recurrent learning for accurate and efficient modeling of sparse spatiotemporal systems.

Accurately modeling and forecasting complex systems governed by partial differential equations (PDEs) is crucial in various scientific and engineering domains. However, traditional numerical methods struggle in real-world scenarios due to incomplete or unknown physical laws. Meanwhile, machine learning approaches often fail to generalize effectively when faced with scarce observational data and the challenge of capturing local and global features. To this end, we propose the Physics-encoded Spectral Attention Network (PeSANet), which integrates local and global information to forecast complex systems with limited data and incomplete physical priors. The model consists of two key components: a physics-encoded block that uses hard constraints to approximate local differential operators from limited data, and a spectral-enhanced block that captures long-range global dependencies in the frequency domain. Specifically, we introduce a novel spectral attention mechanism to model inter-spectrum relationships and learn long-range spatial features. Experimental results demonstrate that PeSANet outperforms existing methods across all metrics, particularly in long-term forecasting accuracy, providing a promising solution for simulating complex systems with limited data and incomplete physics.

The Concept Bottleneck Model (CBM) is an interpretable neural network that leverages high-level concepts to explain model decisions and conduct human-machine interaction. However, in real-world scenarios, the deficiency of informative concepts can impede the model's interpretability and subsequent interventions. This paper proves that insufficient concept information can lead to an inherent dilemma of concept and label distortions in CBM. To address this challenge, we propose the Decoupling Concept Bottleneck Model (DCBM), which comprises two phases: 1) DCBM for prediction and interpretation, which decouples heterogeneous information into explicit and implicit concepts while maintaining high label and concept accuracy, and 2) DCBM for human-machine interaction, which automatically corrects labels and traces wrong concepts via mutual information estimation. The construction of the interaction system can be formulated as a light min-max optimization problem. Extensive experiments expose the success of alleviating concept/label distortions, especially when concepts are insufficient. In particular, we propose the Concept Contribution Score (CCS) to quantify the interpretability of DCBM. Numerical results demonstrate that CCS can be guaranteed by the Jensen-Shannon divergence constraint in DCBM. Moreover, DCBM expresses two effective human-machine interactions, including forward intervention and backward rectification, to further promote concept/label accuracy via interaction with human experts.

In scenarios with limited available data, training the function-to-function neural PDE solver in an unsupervised manner is essential. However, the efficiency and accuracy of existing methods are constrained by the properties of numerical algorithms, such as finite difference and pseudo-spectral methods, integrated during the training stage. These methods necessitate careful spatiotemporal discretization to achieve reasonable accuracy, leading to significant computational challenges and inaccurate simulations, particularly in cases with substantial spatiotemporal variations. To address these limitations, we propose the Monte Carlo Neural PDE Solver (MCNP Solver) for training unsupervised neural solvers via the PDEs' probabilistic representation, which regards macroscopic phenomena as ensembles of random particles. Compared to other unsupervised methods, MCNP Solver naturally inherits the advantages of the Monte Carlo method, which is robust against spatiotemporal variations and can tolerate coarse step size. In simulating the trajectories of particles, we employ Heun's method for the convection process and calculate the expectation via the probability density function of neighbouring grid points during the diffusion process. These techniques enhance accuracy and circumvent the computational issues associated with Monte Carlo sampling. Our numerical experiments on convection-diffusion, Allen-Cahn, and Navier-Stokes equations demonstrate significant improvements in accuracy and efficiency compared to other unsupervised baselines.

Synthetic Aperture Radar (SAR) data, characterized by high spatial resolution and all-weather observation capabilities, holds tremendous potential for application in soil moisture monitoring and crop growth assessment. Currently, the Water Cloud Model (WCM) is widely applied in SAR-based soil moisture inversion. However, the traditional WCM’s reliance on optical vegetation indices, which cannot be synchronized with SAR data, and the incomplete polarization scattering information contained in existing SAR vegetation indices, adversely effect the accuracy of SAR soil moisture retravel. Thus, this paper proposed a novel dual-polarization SAR vegetation index to soil moisture retravel in croplands. Firstly, the method establishes a Polarization Scattering Contrast Parameter (mcp) by combining SAR data covariance elements with backscattering information. Subsequently, based mcp, the Polarization Scattering Correlation Contrast parameter (Rcp) is defined, incorporating the degree of polarization, to integrate both scattering contrast and polarization state characteristics. Then, we proposed a novel dual-polarization SAR vegetation index (DRVIs) based Rcp. Finally, the surface soil moisture retravel in cropland area was realized using DRVIs and multiparameter WCM. This study conducted experimental in the winter wheat and summer corn covered areas of the Eastern Henan Plain, China. The experimental results indicate that during the wheat growing season, the correlation coefficient between the surface soil moisture products and estimated results reached 0.72. And during the corn growing season, the correlation coefficient was approximately 0.64. The research will contribute to expending the application of SAR data in vegetation growth monitoring and soil moisture retrieval.

Surface soil moisture (SSM) is an essential component of surface ecosystems. However, the existing microwave SSM retrieval methods are beleaguered with issues such as insufficient decoupling of terrestrial scattering characteristics and excessive reliance on empirical parameters within the models. This paper proposes a novel generalized SAR backscattering model (GSBM) for time-series retrieval of SSM, considering different land-cover influences. Initially, the GSBM is introduced, categorizing land cover into built-up areas, water bodies, vegetation-covered areas, and soils. Subsequently, we establish a unique SAR Water Cloud Model (SWCM) with the dual-polarization SAR vegetation index (DRVIs). A high-quality soil backscatter coefficient is obtained by employing the SWCM to eliminate vegetation's influence. Ultimately, the dry and wet reference values of soil backscatter are calculated to retrieve the relative SSM time series. Based on Sentinel-1 data, we select the Golmud area for a three-year spatiotemporal monitoring of SSM. Experimental results show that our method improves the correlation coefficients (r) between SAR backscatter and in situ soil moisture data from 0.63 to 0.73, improving about 16%. The r between SSM retrieval results and in situ data was above 0.66 at 500 m, 1 km and 3 km spatial resolutions. Consequently, the proposed method underscores the advantages of simplicity in parameters, high estimation precision, and robust adaptability, thereby augmenting the potential for large-scale global monitoring applications.

Affected by the surrounding environment of the station and GPS satellite signal receiving conditions, there are significant fluctuations in the snow depth retrieval results of the Global Positioning System Interferometric Reflection (GPS-IR) based on GPS data acquired in different trajectories or frequency bands, resulting in the loss of snow depth retrieval accuracy and reliability. Therefore, this contribution proposes a novel snow depth retrieving approach using time-series clustering (TSC) in GPS-IR data. For improving the adaptability of the algorithm to different application scenarios, Dynamic Time Warping (DTW) is introduced to perform K-Medoids clustering on the snow depth retrieval sequence of all available satellite trajectories, and the daily time-series snow depth retrieval results are obtained based on the clustering center evaluation index. For validation purposes, multi-frequency GPS data was selected for snow depth retrieval. The results indicate that, compared with traditional approaches, the proposed TSC method demonstrated higher robustness and accuracy, with correlation coefficients reaching above 0.975 in all frequency bands. The root mean square difference (RMSD) for snow depth retrieval at P351 and AB39 stations, using GPS L1 frequency data, were 8.772 and 6.095cm, respectively. The TSC algorithm proposed in this article classifies the snow depth retrieval sequences of different trajectories, achieving an adaptive acquisition of the optimal snow depth retrieval sequence, thus greatly reducing the workload of manual intervention and preliminary data quality assessment in GPS-IR snow depth retrieval. The proposed GPS-IR snow depth retrieval algorithm may offer a new technical approach for future related research.

Monitoring changes in glacial lake (GL) area is of great significance for revealing climate change and analyzing the risk of GL outburst floods (GLOFs). However, in the case of Southeastern Tibet Plateau (SETP), the availability of optical images in high mountain areas is low, and Synthetic Aperture Radar (SAR) amplitude images have significant geometric distortions. Most GL extraction studies are mainly focused on autumn and winter seasons. In order to obtain accurate GL boundaries at any time and to capture seasonal changes of GLs in a wide area, this letter proposes a high mountain GL extraction framework based on the coupling of Google Earth Engine (GEE) and historical GL catalog data, which focuses on local GL extraction problems. GL extraction and validation were performed using various clustering and adaptive threshold segmentation methods, all of which showed strong stability and reliability of the proposed scheme. Finally, we employed a superpixel clustering algorithm to estimate the area of GLs as of August 1, 2019, and then compared the results with two widely used spatially referenced datasets. The results indicate that our method achieves a comprehensive Intersection over Union (IoU) of up to 95%. The proposed method can effectively support the extraction of wide-area summer GLs and the monitoring of seasonal changes in GL area, thus enabling the dynamic updating of GL information at a high temporal frequency.

Synthetic aperture radar (SAR) is a crucial technique in Earth observation, providing vast amounts of data for monitoring the Earth’s surface. However, SAR’s side-looking imaging characteristics often result in significant geometric distortions in complex terrains such as mountainous gorges. Current methods struggle to accurately compute both active and passive geometric distortions when orbital state vector information is not available. This study aims to address this challenge by concentrating on the Southeastern Tibetan Plateau (SETP) and introducing a DEM-based radar incidence angle-tracking method (Angle-Tracking) based on ray tracing principles. The fundamental aspects of this method include constructing the discrete range direction vector (DRDV) to establish calculation directions, refining grid distribution via cubic-patch cells, and identifying potential topographic occluder points (PTOPs) to minimize redundancy in iterative computations. Through the utilization of this approach, we have acquired and disclosed the distribution of geometric distortion in ascending and descending Sentinel-1 data over the SETP region. Cross validation with results computed from precise orbital data showcases the efficacy of the angle-tracking method in identifying geometric distortions in the absence of satellite state vector information. Furthermore, the angle-tracking method demonstrates effectiveness when implemented on cloud computing platforms such as Google Earth Engine (GEE), thereby enhancing the feasibility of SAR-based research in mountainous areas.

Context-aware methods achieved great success in supervised scene text recognition via incorporating semantic priors from words. We argue that such prior contextual information can be interpreted as the relations of textual primitives due to the heterogeneous text and background, which can provide effective self-supervised labels for representation learning. However, textual relations are restricted to the finite size of dataset due to lexical dependencies, which causes the problem of over-fitting and compromises representation robustness. To this end, we propose to enrich the textual relations via rearrangement, hierarchy and interaction, and design a unified framework called RCLSTR: Relational Contrastive Learning for Scene Text Recognition. Based on causality, we theoretically explain that three modules suppress the bias caused by the contextual prior and thus guarantee representation robustness. Experiments on representation quality show that our method outperforms state-of-the-art self-supervised STR methods. Code is available at https://github.com/ThunderVVV/RCLSTR.

As a supplement to time-series interferometric synthetic aperture radar (TS-InSAR), time-series multiple-aperture InSAR (TS-MAI) can measure the spatiotemporal changes in SAR along-track surface deformation. TS-MAI is often applied with low-frequency SAR data (e.g., L-band data) due to its ability to retain high interferometric coherence. However, the low-frequency SAR signal is vulnerable to ionospheric delays, which can significantly degrade the measurement accuracy of TS-MAI. This letter presents an approach to correct the ionospheric errors in TS-MAI. A polynomial cubic model is employed to constrain the ground deformation, which is then incorporated into the observation model for effectively separating the deformation signal and the ionospheric delays. The proposed method is tested using the L-band ALOS-1 PALSAR-1 datasets covering the Tocopilla area in Chile between November 2007 and March 2011. The correction performance and accuracy of the proposed method are demonstrated by comparing the range split-spectrum interferometry (RSSI)-based method and the local GPS data, respectively. The root mean square error (RMSE) improvement rates between TS-MAI and GPS are 72.17% for the SRGD site and 84.51% for the VLZL site, and their correlation coefficients increase from 0.23 and 0.50 to 0.52 and 0.61 after the correction.

Forest fire not only seriously affects the stability of the forest ecosystem but also threatens the safety of human life and property. The previously burned areas extracting method mainly focuses on optical images, susceptible to cloud and fog objective environmental factors. Although there are related studies on the threshold segmentation of single SAR feature types, further information mining for SAR image data, e.g., backscattering intensity, polarization decomposition, texture, and other features, is still insufficient. Therefore, this letter proposes a burned area extracting method using polarization and texture features of Sentinel-1A images to combine and comprehensively mine various SAR feature change information caused by forest fires with a random forest (RF) model. For validation purposes, we compared the burned areas’ extracted results with the reference data acquired based on Sentinel-2A optical imagery. The comparative results show that the SAR extraction results highly agree with the reference data, with an accuracy of 87.12%, and the commission and omission errors were 20.44% and 12.88%, respectively. The proposed machine learning method helps extract fire areas covered by thick smoke or persistent clouds and provides a reference to related research.

Ionospheric phase delay is a critical error source in Time Series Interferometric Synthetic Aperture Radar (TS-InSAR) for the purpose of monitoring ground surface deformation with SAR data obtained from low-frequency radar systems. Recently, the Range Split-Spectrum Interferometry (RSSI) method has been employed to estimate and rectify ionospheric errors in TS-InSAR. However, the performance of the RSSI method is largely restricted by the significant linear scale factors resulting from the current small SAR bandwidth. In this study, we propose a Reformulating RSSI (Re-RSSI)-based method for correcting the ionospheric error in TS-InSAR by optimizing the linear scale factors, with the aim of improving the accuracy of TS-InSAR measurements. We evaluate the Re-RSSI method using 121 ALOS-1 PALSAR images that cover two distinct regions: the low-latitude Lazufre volcano region and the high-latitude Anaktuvuk River tundra fire region. Our results demonstrate that the Re-RSSI method can effectively remove time series ionospheric errors at both test sites, where we detected ionospheric delays of approximately 2.5 cm/yr and 2.0 cm/yr, respectively. Using Global Navigation Satellite System (GNSS) measurements as ground truth, we achieved an 86.59% improvement rate in root mean square error (RMSE) with the Re-RSSI method, which is significantly higher than the 66.40% improvement rate achieved with the traditional RSSI method.

Understanding the black-box prediction for neural networks is challenging. To achieve this, early studies have designed influence function (IF) to measure the effect of removing a single training point on neural networks. However, the classic implicit Hessian-vector product (IHVP) method for calculating IF is fragile, and theoretical analysis of IF in the context of neural networks is still lacking. To this end, we utilize the neural tangent kernel (NTK) theory to calculate IF for the neural network trained with regularized mean-square loss, and prove that the approximation error can be arbitrarily small when the width is sufficiently large for two-layer ReLU networks. We analyze the error bound for the classic IHVP method in the over-parameterized regime to understand when and why it fails or not. In detail, our theoretical analysis reveals that (1) the accuracy of IHVP depends on the regularization term, and is pretty low under weak regularization; (2) the accuracy of IHVP has a significant correlation with the probability density of corresponding training points. We further borrow the theory from NTK to understand the IFs better, including quantifying the complexity for influential samples and depicting the variation of IFs during the training dynamics. Numerical experiments on real-world data confirm our theoretical results and demonstrate our findings.

Benchmark datasets for visual recognition assume that data is uniformly distributed, while real-world datasets obey long-tailed distribution. Current approaches handle the long-tailed problem to transform the long-tailed dataset to uniform distribution by re-sampling or re-weighting strategies. These approaches emphasize the tail classes but ignore the hard examples in head classes, which result in performance degradation. In this paper, we propose a novel gradient harmonized mechanism with category-wise adaptive precision to decouple the difficulty and sample size imbalance in the long-tailed problem, which are correspondingly solved via intra- and inter-category balance strategies. Specifically, intra-category balance focuses on the hard examples in each category to optimize the decision boundary, while inter-category balance aims to correct the shift of decision boundary by taking each category as a unit. Extensive experiments demonstrate that the proposed method consistently outperforms other approaches on all the datasets.

The pixel offset-tracking (PO) technique developed from correlation matching for use on synthetic aperture radar (SAR) images has been widely employed to monitor glacier dynamics. However, the decorrelation caused by rapid changes in a glacier surface reduces the integrity of flow velocity extraction. In this letter, we propose a novel method, termed the optical flow (OF) small baseline subset(SBAS), developed from the optical flow algorithm, which is defined as the apparent motion of individual pixels on the image plane. The OF algorithm can compute dense flows at the individual pixel level with a low computational cost and may serve as an alternative to PO for estimating the glacier velocity field. During processing, We selected the image pairs having short spatiotemporal baselines and calculated their offset series according to the OF algorithm. We then used an interval estimation strategy to eliminate outliers to refine stacked offsets. Finally, the least-squares method was used to calculate the displacement series for each pixel. We tested the proposed method utilizing eight ALOS-2/PALSAR-2 images on a temperate debris-covered glacier, the Hailuogou Glacier on the southeastern Tibetan Plateau. Compared with PO-SBAS, our method effectively improves the coverage of glacier flow velocity monitoring from 73.5% to 99.6%.

The global positioning system interferometric reflectometry (GPS-IR) method has the advantage of acquiring observations continuously in all weather conditions, which has great application potential in vegetation remote sensing. However, L-band electromagnetic wave signals are susceptible to various environmental factors, resulting in deviations in GPS-IR observation data at certain times. The accuracy and reliability of the vegetation index retrieving results are challenging to achieve by using single-frequency GPS data. This letter proposed a novel method to combine dual-frequency data for retrieving normalized difference vegetation index (NDVI). We integrated the multipath observations based on the theory of information entropy. Subsequently, a unary linear regression model was established to retrieve the NDVI from the calculated normalized microwave reflection index (NMRI). For validation purposes, the comparative analysis was conducted between the proposed model and the previous single-frequency NDVI retrieving model in terms of retrieval accuracy, based on the continuous observation data acquired by four GPS reference stations in the past five years. The experimental results indicated that the proposed model is available for retrieving the NDVI, with the correlation coefficient of 0.749–0.815 and the root mean square error (RMSE) of 0.056–0.081. Compared with the results acquired by the previous single-frequency NDVI retrieving model, the correlation coefficient of the retrieved NDVI was increased by an average of 18.5%, and the RMSE was reduced by 30.3%. The proposed method in this letter helps further improve the accuracy and continuity of NDVI observation data in some local areas, which contributes to grasping the growth status of vegetation comprehensively.

An emerging microwave remote sensing technology, global navigation satellite system interferometry reflectometry (GNSS-IR) shows excellent application potential in vegetation remote sensing due to its all-weather and high temporal resolution characteristics. Previous studies for GNSS-IR have mainly concentrated on the Global Positioning System (GPS) SNR data.Because of the similarity between GPS and BDS (Beidou Navigation Satellite System), the signal of BDS can also be impacted by vegetation, which however, has not been comprehensively researched. Therefore, this letter acquires normalized amplitude of the BDS SNR data based on triple-frequency SNR observations collected by two stations with different vegetation types. To reveal the impact of different vegetation growth conditions on the BDS multi-frequency signals, we conduct a comparative analysis with normalized difference vegetation index (NDVI) data obtained by Sentinel-2 and Moderate-resolution Imaging Spectroradiometer (MODIS) imagery, respectively. The outcomes demonstrate that the normalized amplitude of the BDS signal exhibits a strong association with Sentinel-2 NDVI, with correlation coefficients varying from 0.69 to 0.83 and 0.78 to 0.84 at P041 and P105, respectively. In addition, it is the first time discovered that harvesting vegetation around the stations leads to a cliff-like decrease in normalized amplitude. Compared with optical remote sensing vegetation monitoring methods, BDS-IR shows significant temporal resolution and sensitivity advantages. This finding may promote vegetation monitoring, vegetation protection, and other related research fields.

Multiple aperture interferometric synthetic aperture radar (MAI) can measure ground displacements along SAR track. However, MAI measurements may suffer from severe ionospheric error, especially for long-wavelength SAR sensors. This study presents a new approach, azimuth split-spectrum interferometry (AziSSI), to correct ionospheric errors in MAI measurements. The proposed method can straightforwardly resolve ionospheric phase delay by exploiting two subband MAI interferograms with different centroid frequencies. We utilized two groups of ALOS-1 PALSAR images, which cover the 2008 Wenchuan earthquake containing strong coseismic ground deformation and a stable coast region in Chile, respectively, to test the proposed method. Experimental results show that the AziSSI method can successfully recover the coseismic displacements induced by the Wenchuan earthquake and circumvent the azimuth stripes for the Chile case’s MAI measurements. The maximum ionosphere-induced azimuth shifts are about 2.2 m for the Wenchuan case and 2.5 m for the Chile case. We validated the correction performance of the AziSSI method with the Wenchuan case by comparing the compensated azimuth displacements with the simulated deformation from a forward fault slip model and coseismic global positioning system (GPS) measurements. The validations show that the deformation patterns after the ionospheric correction are consistent with the simulations of the sinistral slip of the causative fault. The root-mean-square error between the GPS and MAI measurements is reduced from 0.77 to 0.28 m before and after the ionospheric correction, indicating that AziSSI can significantly compensate ionospheric errors for MAI.

Link prediction and network alignment are two fundamental and interleaved tasks in network analysis. In this paper, we propose a novel cross-network embedding model under the Skip-gram framework, which alternately performs link prediction and network alignment by joint optimization. Vertex sequences, obtained via a biased random walk based on empirical mixture distributions, are used to train a Skip-gram based node embedding model. On one hand, based on the similarity in embedding space, network alignment can be effectively performed either with the initial ground truth alignments as seeds or from scratch. On the other hand, the proposed link prediction model involves training a supervised classifier by sampling a set of positive and negative edges. We also modify and incorporate the Collective Link Fusion (CLF) method under a Skip-gram framework and show that the new method can achieve better results in both tasks. Extensive experimental results show the state-of-the-art performance of our methods.

Mounting evidence has shown the involvement of long non-coding RNAs (lncRNAs) during various cancer metastatic events (abbreviated as CMEs, e.g. cancer cell invasion, intravasation, extravasation, proliferation, etc.) that may cooperatively facilitate malignant tumor spread and cause massive patient deaths. The study of lncRNA-CME associations might help understand lncRNA functions in metastasis and present reliable biomarkers for early dissemination detection and optimized treatment. Therefore, we developed a database named 'lncR2metasta' by manually compiling experimentally supported lncRNAs during various CMEs from existing studies. LncR2metasta documents 1238 associations between 304 lncRNAs and 39 CMEs across 54 human cancer subtypes. Each entry of lncR2metasta contains detailed information on a lncRNA-CME association, including lncRNA symbol, a specific CME, brief description of the association, lncRNA category, lncRNA Entrez or Ensembl ID, lncRNA genomic location and strand, lncRNA experiment, lncRNA expression pattern, detection method, target gene (or pathway) of lncRNA, lncRNA regulatory role on a CME, cancer name and the literature reference. An easy-to-use web interface was deployed in lncR2metasta for its users to easily browse, search and download as well as to submit novel lncRNA-CME associations. LncR2metasta will be a useful resource in cancer research community. It is freely available at http://lncR2metasta.wchoda.com.

The persistent scatterer interferometric synthetic aperture radar (PSI) has proven to be a powerful tool for monitoring surface deformation. However, the conventional deformation models cannot be fully adapted to analysis of the complicated deformation process. Taking into account seasonal response and tectonic movement, this letter presents an integrated deformation model for separating deformation components through PSI time series, thus obtaining the linear deformation rate, the deformation acceleration and the seasonal deformation amplitude at each PS. An iterative solution method is proposed to estimate the parameters in the integrated model. The experiments are carried out for subsidence detection over the northwestern part of Tianjin (China) by using the 40 high-resolution TerraSAR-X SAR images acquired between 2009 and 2010 and the ground truth data collected by precise leveling at seven benchmarks and six man-made corner reflectors. The testing results indicate that the integrated model has better adaptability to the subsidence process in the study area than the conventional models. The further analysis shows that the deformation results derived from the iterative solution are in better agreement with the ground truth data. These demonstrate that the proposed methodology is effective for monitoring the surface deformation with remarkable nonlinear property.

In this paper, we present an improved approach of the multitemporal interferometric synthetic aperture radar (InSAR) for detecting land subsidence in coastal areas by using the time series of high-resolution SAR images. In particular, our algorithm extends the capability of the temporarily coherent point InSAR (TCPInSAR) technique that can be used to detect subsidence even in the case of a small number of SAR images available for a study area. The proposed approach is implemented by using the interferograms with ultrashort spatial baselines (USBs) through several procedures, including the selection of USB interferometric pairs, TCP identification, TCP networking and modeling, as well as TCP solution by a least squares estimator. As the topographic effects in coastal areas are negligible in the USB interferograms, an external digital elevation model is no longer necessary for differential processing, thus simplifying both TCP modeling and parameter estimating. The USB-based TCPInSAR algorithm has been tested with the high-resolution TerraSAR-X images acquired over Tianjin (close to Bohai Bay) in China, and validated by using the ground-based leveling measurements. The testing results indicate that the density and coverage extent of TCPs can be increased dramatically by using the proposed algorithm, and the quality of subsidence measurements derived by the USB-based TCPInSAR can be raised.

Recent coding-based image classification systems generally adopt a key step of spatial pooling operation, which characterizes the statistics of patch-level local feature codes over the regions of interest (ROI), to form the image-level representation for classification. In this paper, we present a hierarchical ROI dictionary for spatial pooling, to beyond the widely used spatial pyramid in image classification literature. By utilizing the compositionality among ROIs, it captures rich spatial statistics via an efficient pooling algorithm in deep hierarchy. On this basis, we further employ partial least squares analysis to learn a more compact and discriminative image representation. The experimental results demonstrate superiority of the proposed hierarchical pooling method relative to spatial pyramid, on three benchmark datasets for image classification.

In this paper, we propose a new algorithm for image re-ranking in web image search applications. The proposed method focuses on investigating the following two mechanisms: 1) Visual consistency. In most web image search cases, the images that closely related to the search query are visually similar. These visually consistent images which occur most frequently in the first few web pages will be given higher ranks. 2) Visual saliency. From visual aspect, it is obvious that salient images would be easier to catch users' eyes, and it is observed that these visually salient images in the front pages are often relevant to the user's query. By integrating the above two mechanisms, our method can efficiently re-rank the images from search engines and obtain a more satisfactory search result. Experimental results on a real-world web image dataset demonstrate that our approach can effectively improve the performance of image retrieval.

The paper investigates two mechanisms, visual consistency and visual saliency, in web image search: (1) In most current web image search engines, such as Google Image Search and Yahoo Image Search, the images that closely related to the search query are typically visually similar. These visually consistent images which occur most frequently in the first few web pages will be given higher ranks. (2) From visual aspect, it is obvious that salient images would be easier to catch users' eyes and more likely to be clicked than the cluttered ones in low-level vision. In addition, we also observe the fact that the visually salient images in the front pages are often relevant to the user's query. The principal novelty of this paper is in combining visual saliency and consistency to re-rank the results from search engines to make the re-ranked images more satisfying in both vision and content. The experimental results on a real world web image dataset demonstrate that our approach can effectively improve the performance of image retrieval.

In this paper we present a complete framework for scene categorization that builds upon and extends several recent ideas including spatial pyramid representation and a variety of base local descriptors which have different discriminative power and invariance from task to task. Furthermore, we propose two strategies: sum-max and max-max, used to effectively combine diverse source of data in a unified setting way. Our approach shows significantly improved performance on a large, challenging data set of fifteen natural scene categories. Owing to combination of complementary information cues, our approach is expected to equally applicable to a range of tasks.

Traditional image interpolation methods assume that the local spatial structure of the low-resolution (LR) and high-resolution (HR) images are approximately the same, and use edge information of the LR image to estimate the missing pixels. This assumption, however, no longer holds for natural images with fine and dense textures. Consequently, those methods cannot restore dense textures well and tend to generate over-fitting visual effects. In this paper, a learned HR image prior is exploited to overcome the problems. In particular, we use Fields of Experts (FoE) with student's t-distribution experts to model the prior, taking advantage of its representative ability of non-Gaussian natures in images. Then Maximum a Posterior (MAP) estimation incorporating FoE prior is used to estimate the missing pixels. Experimental results compared with traditional interpolation methods demonstrate that our method not only can recover fine details and produce superior PSNR values, but also avoid the visual over-fitting problems.

In learning based single image super-resolution (SR) approach, the super-resolved image are usually found or combined from training database through patch matching. But because the representation ability of small patch is limited, it is difficult to guarantee that the super-resolved image is best under global view. To tackle this problem, we propose a statistical learning method for SR with both global and local constraints. Firstly, we use maximum a posteriori (MAP) estimation with learned image priors by fields of experts (FoE) model, and regularize SR globally guided by the image priors. Secondly, for each overlapped patch, the higher-order Markov random fields (MRFs) is used to model its local relationship with corresponding high-resolution candidates, then belief propagation is used to find high-resolution image. Compared with traditional patch based learning method without global constraint, our method could not only preserve the global image structure, but also restore the local details well. Experiments verify the idea of our global and local constraint SR method.

In this paper we propose PHOTO (pyramid histogram of topics), a new representation for image classification. We partition the image into hierarchical cells and learn the topic histogram using pLSA over each cell with EM algorithm. Then we concatenate the topic histograms over the cells at all levels to form a ldquolongrdquo vector, i.e. pyramid histogram of topics. Finally AdaBoost classifiers are used to select the topics most discriminative for class recognition. Experimental results on two diverse databases show that our method performs significantly better than general topic representation.

We propose a new approach for recognizing object classes which is based on the intuitive idea that human beings are able to perform the task well given only thumbnails (coarse scale version) of images. Unlike previous work which uses local image features at fine scales, our approach uses thumbnails directly, and captures their high-order correlations at coarse scales through deep multi-layer neural networks based on restricted Boltzmann machines. Specifically, the pretraining stage of such networks takes on the role of feature extraction. Experimental results show that the proposed approach is comparable to other state-of-the-art recognition methods in terms of accuracy. The merits of the proposed approach come from the simplicity of the workflow and the parallelizability of the implementation structure.

In patch based face super-resolution method, the patch size is usually very small, and neighbor patchespsila relationship via overlapped regions is only to keep smoothness of reconstructed high-resolution image, so the prior is not always strong enough to regularize super-resolution when observed low-resolution image lose facial structure information. We propose to use Gaussian Mixture Model(GMM) to learn facial prior embedded between un-overlapped regions of neighbor patches. This approach, which has never been used to regularize face super-resolution before, usually works as a potential function in 8-connected Markov Random Fields (MRFs) with belief propagation. In the proposed algorithm, we assign high probability to the neighbor candidate patches that express correct facial structure, and others not. Experiments demonstrate that our method is superior in preserving smoothness and recovers facial structure and local details when low-resolution image lost the details of facial structure.