High-resolution remote sensing images (HRRSI) have important theoretical and practical value in urban planning. However, current segmentation methods often struggle with issues like blurred edges and loss of detailed information due to the intricate backgrounds and rich semantics in high-resolution remote sensing images. To tackle these challenges, this paper proposes an end-to-end attention-based Convolutional Neural Network (CNN) called Double Hybrid Attention U-Net (DHAU-Net). We designed a new Double Hybrid Attention structure consisting of dual-parallel hybrid attention modules to replace the skip connections in U-Net, which can eliminate redundant information interference and enhances the collection and utilization of important shallow features. Comprehensive experiments on the Massachusetts remote sensing building dataset and the Inria aerial image labeling dataset demonstrate that our proposed method achieves effective pixel-level building segmentation in urban remote sensing images by eliminating redundant information interference and making full use of shallow features, and improves the segmentation performance without significant time costs (approximately 15%). The evaluation metrics reveal significant results, with an accuracy rate of 0.9808, precision reaching 0.9300, an F1 score of 0.9112, a mean intersection over union (mIoU) of 0.9088, and a recall rate of 0.8932.

Spatiotemporal traffic data is vitally important in intelligent transportation systems, but traffic data values are inevitably missing due to malfunctions in sensing devices or transmission networks. This will cause trouble for the subsequent traffic flow prediction work. Low-rank tensor imputation methods have recently received increasing attention to traffic data imputation due to their ability to capture the spatial-temporal information in the way of multi-dimension. However, traffic data often encounters challenging multimodal missing scenarios, including random missing, fiber missing, and slice missing. Thus, only accounting for the global low-rankness of traffic data is less sufficient to recover missing data. Motivated by the local smoothness of traffic data, we introduce the hybrid total variation into spatiotemporal traffic data imputation framework. The proposed model consists of the tensor nuclear norm, the first-order and the second-order total variation, which presents the global low-rankness and local smoothness prior, respectively. We utilize the ADMM to effectively solve the proposed model and establish the theoretical guarantee. Numerous experiments on two real-world datasets under three multimodal missing scenarios indicate that we proposed method outperforms some state-of-the-art traffic data imputation methods, especially for fiber missing and slice missing.

Spatialtemporal traffic data are collected by a wide array of data collection devices deployed in intelligent transportation systems, which play crucial roles in data-driven intelligent transportation systems. However, hardware malfunctions or software system errors can lead to the inability to collect accurate data. Incomplete data in intelligent transportation systems can cause difficulties for subsequent applications, such as traffic flow anomaly detection and forecasting. Recently, low-rank tensor-based methods that characterize the global correlation of high-dimensional data have achieved superior performance. However, the majority of tensor-based approaches primarily account for the global correlation within the target traffic data, which is not sufficient to recover some challenging missing scenarios, such as fiber missing and slice missing that often occur in real situations. In this manuscript, we suggest a flexible method to explore the global correlation, local smoothness, and nonlocal self-similar redundancy in traffic data to improve recovery accuracy. Specifically, we utilize the multidimensional tensor nuclear norm to characterize the correlation structure, the multidimensional total variation to maintain smoothness detail, and the plug-and-play term to promote nonlocal self-similarity. The primary merits of this model lie in the fact that these priors characterize data features from different perspectives and complement each other. We develop an alternating direction method of multipliers to achieve efficient optimization for each variable. The proposed method is comprehensively evaluated through experiments across seven missing scenarios. Extensive experiments demonstrate that the suggested method with the aid of different priors outperforms many state-of-the-art approaches, especially structured missing scenarios.

Since the facility restrictions and weather conditions, hyperspectral image (HSI) is generally seriously polluted by a variety of noises. Recently, the method based on block-term decomposition with rank-(L, L, 1) (BTD) has attracted wide attention in HSI mixed noise removal. BTD factorizes third-order HSI data into the sum of a series of component tensors, where each of the component tensors is represented by the outer product of a rank-L matrix A _r B _r ^T and a column vector c _r . BTD has clear physical interpretation because its latent factors A _r B _r ^T and c _r can be interpreted abundance map and spectral signature, respectively. The essential uniqueness of BTD is under the low-rank assumption of A _r B _r ^T . However, the low-rank assumption is not always held in real scenarios. The BTD-based method usually sets L to full rank to achieve satisfactory results. In this article, we suggest a novel model based on nonlocal block-term decomposition (NLBTD) for HSI mixed noise removal. More specifically, for each grouped similar image block, BTD is introduced to capture nonlocal self-similarity and global spectral low-rankness, the unidirectional total variation is introduced to preserve local spectral smoothness. By faithfully exploring nonlocal self-similarity, global spectral low-rankness, and local spectral smoothness, the proposed method is expected to produce promising results with guarantee the essential uniqueness of BTD. To tackle the resulting model, we design an efficient algorithm based on the proximal alternating minimization with the theoretical guarantees. Extensive numerical experiments in HSI mixed noise removal demonstrate that the proposed NLBTD method achieves satisfactory performance compared with state-of-the-art methods.

Seismic data are inevitably polluted by noise, which has a significant impact on subsequent applications. Currently, regularization-based methods that utilize interpretable data priors are widely used for eliminating seismic data noise. However, many existing methods are merely direct linear weighted combinations of regularization terms, resulting in the introduction of a series of balancing parameters. Moreover, these methods may fail to fully capture the complex curved structures that exist in seismic data. To handle the above problems, we propose a merged approach based on nonlocal enhanced low-rankness and second-order gradient smoothness (ELGS). The ELGS method simultaneously encodes multiple dimensional features of seismic data within a single term, including temporal evolution characteristics, geological spatial structural information, and self-repeating redundancy of seismic profiles, which minimizes the introduction of balancing parameters while effectively capturing the curved geological structures of seismic data. Specifically, the ELGS method maps low-dimensional seismic data to high-dimensional structural tensors by using a similar patch matching framework and applies low-rank constraints to the second-order gradient tensors along each direction for each high-dimensional structural tensor. To solve the optimization problem faced by the developed ELGS approach, we design an algorithm based on the alternating direction multiplier method. Numerous numerical experiments conducted on both synthetic and field seismic data show that our approach outperforms the related approaches in both quantitative measures and visual performance.

Urban vitality (UV) is the core index to measure the quality and sustainability of urban development. Accurately analyzing the complex association mechanism between UV and built environment (BE) is critical to urban planning practice. Focusing on the main urban area of Chengdu, this study integrates eight categories of multi-source data, including nighttime light data, WorldPop population distribution data, street view images, and POI data, to construct a four-dimensional UV evaluation system and identify 26 BE factors. Firstly, the UV level is quantified by objective weighting methods. Secondly, an XGBoost model combined with a SHAP framework is adopted to investigate the nonlinear association between UV and BE factors. Finally, a spatial autocorrelation model, SHAP spatial visualization and clustering methods are employed to reveal the spatial pattern of UV and the spatial heterogeneity of the association between UV and BE. The results indicate: (1) Various elements of the BE show a significant nonlinear association and threshold effect for UV. Catering services and public transit services are the core factors for UV prediction, with their combined contribution accounting for 37.47%. (2) UV shows obvious spatial differentiation and agglomeration characteristics. It presents a spatial pattern with a gradual decline from the core to the periphery. (3) The association between UV and BE presents spatial heterogeneity, and the predictive contribution logic differs distinctly across different concentric rings. The study conclusions provide a scientific basis for UV improvement and BE optimization in Chengdu.