The Internet of Things (IoT) enables large-scale environmental monitoring, yet sparse and irregular sensor layouts make it difficult to reconstruct fine-scale spatiotemporal (ST) fields. This paper addresses online super-resolution of high-dimensional environmental fields under sparse IoT sensing by proposing a hierarchical incremental ST perception framework. First, an interpretable low-rank ST decomposition extracts orthonormal spatial modes and compact temporal coefficients to guide sensor placement and model updating. Second, an information-optimal sparse sensing strategy selects sensor subsets by maximizing information gain while enforcing spatial coverage and anchor constraints. Third, a hierarchical incremental learner updates the spatial subspace on the Grassmann manifold and predicts temporal coefficients using a Lyapunov-stable radial basis function (RBF) network, enabling continual adaptation without full retraining. Experiments on a 61-year Pacific Ocean temperature dataset show that the framework achieves over 100-fold spatial super-resolution with 98.46% accuracy, outperforming state-of-the-art baselines under identical sensor budgets and offering an efficient solution for dynamic and sparse IoT perception.

Multi-view stereo (MVS) implicitly encodes photometric and geometric cues into the cost volume for multi-view correspondence matching, transferring insufficient geometric cues essential to depth estimation and reconstruction. This paper proposes GE-MVS, a novel multi-view stereo network with geometric encoding for more accurate and complete depth estimation and point cloud reconstruction. First, the cross-view adaptive cost volume aggregation module is proposed to strengthen multi-view geometric cues encoding during cost volume construction. Then, the depth consistency optimization is performed in the 3D point space during learning by invoking ground-truth depth cues from adjacent views. Finally, the surface normal geometries are explicitly encoded to refine the sampled depth hypotheses to be consistent in the local neighbor regions. Extensive experiments on the standard MVS benchmarks including DTU, Tanks and Temples, and BlendedMVS demonstrate the state-of-the-art depth estimation and point cloud reconstruction performance of GE-MVS. The GE-MVS is further deployed in real-world experiments for UAV-based large-scale reconstruction, where our method outperforms the prevalent industrial reconstruction solutions concerning reconstruction efficiency and efficacy. Our project page is: https://cuhk-usr-group.github.io/GE-MVS/

Recent advancements in learning-based multi-view stereo (MVS) have demonstrated significant improvements over traditional counterpart, primarily due to the extensive availability of multi-view training images with ground-truth metric depths in the terrestrial in-air domain. However, underwater multi-view stereo (UwMVS) faces substantial challenges arising from the domain gap between in-air and underwater environments, leading to degraded performance when applying in-air MVS models to underwater scenarios. Furthermore, the progress of learning-based UwMVS methods has been hindered by the scarcity of underwater multi-view images with ground-truth depth maps and point clouds. In this paper, we address these challenges by introducing a physically-guided approach for synthesizing underwater multi-view images and present the first large-scale UwMVS dataset for end-to-end training and evaluation of learning-based UwMVS methods. Furthermore, we propose a novel UwMVS network that enhances geometric cue encoding to achieve more accurate and complete point cloud reconstruction. Extensive experiments on our dataset and real-world underwater scenes demonstrate that our dataset enables the trained models for underwater dense reconstruction and that our method achieves state-of-the-art performance in underwater reconstruction. Dataset, code and appendix are available at: https://cuhk-usr-group.github.io/UwMVS/

Defect diagnosis in urban infrastructure is crucial for public safety. Traditional manual inspections face significant challenges in terms of accuracy and cost-effectiveness. In this paper, we propose a lightweight and hardware-friendly large-scale infrastructure detector, CUPID, highly suitable for unmanned aerial vehicles (UAVs). Given the significant challenges in automatically detecting defects of varying intensity and size within complex infrastructure, along with the tendency of lightweight models to lose detail and fail to fully capture features during the defect extraction process, we propose the CUPID_Block, a multi-level information fusion block to construct the backbone, featuring the CUPID_Conv module equipped with our proposed CCA (CrissCross Attention). Furthermore, CUPID features an auxiliary training branch that assimilates lower feature maps, helping to recover details lost in deeper convolutional layers. To verify the effectiveness of CUPID and to address the lack of a suitable dataset in the community, we establish a multi-scenario infrastructure defect dataset, CUBIT2024, to conduct extensive experiments. Finally, to assess the efficiency and adaptability of CUPID in UAV for online infrastructure inspection, we design a compact autonomous drone, CU-Astro, where the proposed CUPID is deployed on the Jetson Orin NX computer onboard to evaluate the speed and power consumption of the inference.

Accurately predicting ocean subsurface temperature is vital for advancing ocean and climate research, particularly given the sparse and costly nature of subsurface observations. This study introduces sparse-to-dense prediction of ocean subsurface temperature using multi-level spatiotemporal (ST) information fusion. The framework integrates interpretable ST decoupling, adaptive feature updating, and sparse-to-dense information fusion modules to address the challenge of sparse observations and ever-evolving dynamic environments. Comprehensive experiments focused on the Pacific demonstrate the superiority of the proposed methodology over peer methods. The proposed methodology achieves high-resolution predictions with a root mean square error of 0.2230, accuracy of 0.9846, and point-wise prediction errors below 0.5°C under 10% online random sparse observations (ORSO). Analyses of spatial and temporal temperature dynamics reveal long-term warming trends in the Pacific, including a temperature rise of up to 2.8°C at -100 m in low-latitude regions over the past 40 years, and identify the latitudinal slope of thermocline dynamics. This study advances the understanding of multi-scale thermal processes and variability in the Pacific, demonstrating the potential of application in climate studies, marine resource management, and environmental monitoring.

Predicting subsurface temperatures is critical for comprehending ocean dynamics and climate shifts. This study presents an incremental stable/dynamic (SD) disentanglement learning framework merging data-driven methods with physics-based insights. It separates stable and dynamic temperature modes to untangle intricate spatiotemporal interactions. To accommodate ongoing data influx, we introduce a recursive evolution approach for updating stable representations, employing orthogonal-triangular decomposition (QR) to capture incremental information. Moreover, a retrospective learning algorithm, guided by temporal changes and ocean temperature correlations, is employed to track the dynamic behavior adaptively. The subsurface temperature fields can be efficiently reconstructed and predicted after model convergence. Extensive experiments validate the model across various depths (−2.5 to −800 m) and times (from May 1964 to December 2021), achieving robust performance metrics: root mean square error (RMSE) of 0.1362, mean absolute error (MAE) of 0.0901, accuracy (ACC) of 0.9911, and coefficient of determination ($R^{2}$) between predictions and observations of 0.9998. Comparative analysis underscores the proposed method’s interpretability, adaptability, and overall performance superiority. Temperature anomaly analysis accurately identifies subsurface decadal oscillations in the low- and mid-latitude Pacific regions.

Underwater optical remote sensing is crucial for geoscience applications but often suffers from image degradation due to complex underwater environments. While learning-based methods have advanced underwater image enhancement (UIE), their efficacy in real-world UIE applications still faces challenges. This limitation arises from training predominantly on synthetic underwater images, resulting in a significantinter-domain gap when applied to real-world data. Additionally, diverse underwater conditions introduceintra-domain challenges, such as color casts and haze, further complicating the UIE process. To address these issues, we propose SSD-UIE, a semi-supervised domain-adaptive framework designed to mitigate bothinter- andintra-domain gaps. Our approach employs a systematic synthesis pipeline to reduce visualinter-domain discrepancies and introduces a Large Synthetic-Real Underwater Image Dataset (LSRUID) to facilitate the training of the framework. The Semantic-Blender is developed to handle semanticinter-domain differences, while the Intra-domain-aware Feature Extraction (IFE) branch and feature alignment strategy effectively addressintra-domain variability. Furthermore, the Dual-Trans Block is introduced to enhance the UIE performance while maintaining computational efficiency. Extensive experiments demonstrate that SSD-UIE outperforms state-of-the-art (SOTA) UIE methods in both qualitative and quantitative evaluations on real-world underwater images. Codes and dataset will be publicly available at https://github.com/RockWenJJ/SSD-UIE.git.

Multi-view stereo (MVS) enables accurate and complete 3D reconstruction from multi-view imagery, serving as a core methodology in remote sensing applications across terrestrial and underwater domains. Recent advancements in learning-based MVS have demonstrated significant improvements over traditional counterparts, primarily due to the extensive availability of multi-view training images with ground-truth metric depths in the terrestrial in-air domain. However, underwater multi-view stereo (UwMVS) faces substantial challenges arising from the domain gap between in-air and underwater environments, leading to degraded performance when applying in-air MVS models to underwater scenarios. Furthermore, the progress of learning-based UwMVS methods has been hindered by the scarcity of underwater multi-view images with ground-truth depth maps and point clouds. In this paper, we address these challenges by introducing a physically-guided approach for synthesizing underwater multi-view images and presenting the first large-scale synthetic UwMVS dataset preserving real-world underwater degradation properties for end-to-end training and evaluation of learning-based UwMVS methods. Furthermore, we propose a novel UwMVS network that enhances geometric cue encoding to achieve more accurate and complete point cloud reconstruction. Extensive experiments on the dataset and real-world underwater scenes demonstrate that our dataset enables the trained models for underwater dense reconstruction and that our method achieves state-of-the-art performance in underwater reconstruction. Dataset, appendix, and supplementary video are available at https://yang-sober.github.io/UnderMVS/.

In printed circuit board (PCB) production, uncertainty in the manufacturing stage and sampling uncertainty in the inspection stage have a great impact on product quality. This article proposes rotating angles based on the optimal detection of PCB defects under uncertainties. First, the principal features are extracted and factors of information concentration are designed to construct the reference template from the preprocessed clean dataset. After that, a rotation-angle-based feature extraction and concentration optimization are performed to select the better representative angles. Then, all optimized features are probabilistically synthesized and compared with the reference template for defect detection. Finally, extensive experiments are conducted on the bare PCB and high-resolution integrated PCB to demonstrate the proposed method is efficient and faster and requires fewer model parameters.

Marine intelligent transportation systems (M-ITS) face challenges in achieving adaptability and efficiency under dynamic spatiotemporal (ST) environments. This paper proposes a novel multimodal underwater transformable agent (MUTA) designed as an adaptive and efficient information node for M-ITS. The MUTA integrates a biomimetic morphing wing, multi-drive propulsion system, incremental environmental perception, and an uncertainty-aware dynamic modeling framework to switch smoothly among long-range (LR) cruising, high-maneuverability (HM) operation, and synergy modes. Comprehensive lake and sea trials demonstrate that the MUTA maintains high motion precision and achieves significant energy efficiency improvements, reaching depths up to 1200 m and covering a range of over 3000km. The MUTA has achieved over 17% efficiency gain in lake and marine environments. This work provides an adaptive and efficient solution for M-ITS under uncertain marine conditions.

Underwater robots are critical observation platforms for diverse ocean environments. However, existing robotic designs often lack long-range and deep-sea observation capabilities and overlook the effects of environmental uncertainties on robotic operations. This paper presents a novel long-range underwater robot for extreme ocean environments, featuring a low-power dual-circuit buoyancy adjustment system, an efficient mass-based attitude adjustment system, flying wings, and an open sensor cabin. After that, an extended environment perception strategy with incremental updating is proposed to understand and predict full hydrological dynamics based on sparse observations. On this basis, a real-time dynamic modeling approach integrates multibody dynamics, perceived hydrological dynamics, and environment-robot interactions to provide accurate dynamics predictions and enhance motion efficiency. Extensive simulations and field experiments covering 600 km validated the reliability and autonomy of the robot in long-range ocean observations, highlighting the accuracy of the extended perception and real-time dynamics modeling methods.

Defect detection is an essential part of quality management for industrial processes. Existing vision-based detection methods are inefficient when uncertainty exists in the industrial image. This paper proposes a systematic methodology for defect detection of uncertain industrial images. It consists of adaptive convolution and confidence learning. First, a convolution model adaptively fuses multiple kernel prediction results, which is employed to learn image defect variation. After that, a confidence learning method is developed to filter the label noise and fine-tune the adaptive convolution model. Finally, experimental studies indicate the proposed method can achieve satisfactory detection accuracy and robustness.