The core task of semantic segmentation is to assign predefined category labels to each pixel in an image, thereby distinguishing between different objects and backgrounds. Few-shot semantic segmentation (FSS) is a specialized semantic segmentation task that aims to accurately segment pixel-level targets of novel classes in query images, relying only on a limited number of annotated support samples to enable rapid adaptation to unseen categories without extensive labeled data. FSS in remote sensing imagery is a critical yet challenging task, primarily due to two intrinsic data characteristics: extreme scale variations among target objects and significant intraclass heterogeneity. These challenges severely degrade the performance of existing FSS methods, which often rely on single, global prototypes and are not explicitly designed for such variability. To address these limitations, we propose GlocalDualNet, a novel FSS framework tailored for remote sensing applications. GlocalDualNet integrates two core technical contributions. First, a multiscale support prototype extraction module generates a set of heterogeneous local prototypes in addition to a conventional global prototype. This approach mitigates the spatial detail loss associated with global-only representations and provides a more comprehensive feature signature for matching. Second, a dual-branch segmentation network is designed to explicitly disentangle the feature learning process for large- and small-scale targets, thereby improving segmentation accuracy across disparate scales. Experimental validation on the iSAID-5<sup>i</sup> benchmark dataset demonstrates that our proposed modules yield a notable 2.13% improvement in segmentation accuracy, establishing the efficacy of the GlocalDualNet framework.
Synthetic aperture radar (SAR) encounters distinct challenges in airborne surveillance (dynamic scene variations, target edge blurring) and spaceborne observation (large-scale analysis, high-resolution processing). Both traditional methods and contemporary deep learning-based solutions exhibit limitations: inadequate dynamic target adaptability, weak small-target detection, and redundant recognition in large-scale scenarios, stemming from challenges like target ambiguity, occlusion, and interclass similarity. To address these challenges, we propose the regional spatial-channel information weighted attention network. The innovations encompass the following. 1) Regional spatial channel attention integrates regional weighting in spatial attention (SA) to amplify key positional features while suppressing speckle noise and edge weak samples. Channel self-attention enhances cross-regional interactions to capture target-environment scattering correlations. 2) Boundary-aware loss employs edge overlapping penalties to improve localization of fuzzy shadow edges, with adaptive weighting to amplify small-target gradient contributions during backpropagation. 3) Context-preserving sliding window detection strategy for large-scale images, which can carry out comprehensive and robust detection. Experimental results demonstrate state-of-the-art performance, with the <italic>m</italic>AP<sup>50</sup> of 99.35% on Sandia National Laboratories video SAR dataset, 97.50% on MSAR-1.0 dataset, and superior large-scale detection capability on MSAR-1.0 and LS-SSDDD datasets.
Synthetic aperture radar (SAR) ship classification is crucial for maritime surveillance. Most existing methods primarily focus on visual or polarimetric features, often constrained by a limited feature set and facing challenges in data diversity and multimodal information integration. This study introduces a text-enhanced multimodal framework for SAR ship classification (TeMSC), an extensible and unified approach that integrates multimodal information related to SAR ships. It consists of text-form geometry information embedding, polarization and visual information embedding, and a multimodal prediction module. By incorporating ship geometry information in text format, TeMSC leverages text representation to enhance feature expressiveness, compensating for the limited discriminative power of traditional visual and polarization features, especially in low-resolution scenarios. TeMSC effectively processes complementary multimodal information through a multimodal prediction module, while avoiding the complexity associated with traditional decision-level feature fusion strategies. In addition, a classification token mechanism is introduced to streamline the classification process. Through a two-stage training strategy, TeMSC captures information across multiple SAR datasets, enhancing its generalization and adaptability. Extensive experiments on the FUSAR-Ship and OpenSARShip datasets demonstrate the superior performance of TeMSC and highlight the benefits of multimodal integration for SAR ship classification. TeMSC provides a foundation for future research on SAR-focused multimodal learning applications.
Nastaran Saberi, Mohammad Hossein Shaker, Claude R. Duguay
et al.
This article presents a method to improve the usability of lake ice cover (LIC) maps generated from moderate resolution imaging spectroradiometer (MODIS) top-of-atmosphere reflectance data by providing estimates of aleatoric and epistemic uncertainty. We used a random forest (RF) classifier, which has been shown to have superior performance in classifying lake ice, open water, and clouds, to generate daily LIC maps with inherent (aleatoric) and model (epistemic) uncertainties. RF allows for the learning of different hypotheses (trees), producing diverse predictions that can be utilized to quantify aleatoric and epistemic uncertainty. We use a decomposition of Shannon entropy to quantify these uncertainties and apply pixel-based uncertainty estimation. Our results show that using uncertainty values to reject the classification of uncertain pixels significantly improves recall and precision. The method presented herein is under consideration for integration into the processing chain implemented for the production of daily LIC maps as part of the European Space Agency's Climate Change Initiative (CCI+) Lakes project.
Aina Garcia-Espriu, Estrella Olmedo, Veronica Gonzalez-Gambau
et al.
The soil moisture and ocean salinity (SMOS) satellite mission, launched in 2009, provides global measurements of sea surface salinity (SSS) using L-band radiometry. In this article, we revisit the algorithms to empirically correct the residual latitudinal and seasonal biases seen in the debiased non-Bayesian (DNB) retrieval algorithm. We characterize these biases that affect the retrieved SSS and derive empirical corrections to mitigate them. We also revisit the filtering criteria for the new release of brightness temperature (v724). We compare the SSS retrieved by using different DNB filtering configurations with in-situ Argo measurements and discuss the importance of these corrections and the optimal configuration parameters for the generation of the Level 3 SSS maps. We observe significant improvements in the retrieved SSS when compared to the ones retrieved in the Barcelona Expert Center global SSS product v2.
Sixth-generation (6G) networks, offering ultra-low latency and high bandwidth, provide critical support for rapid data transmission in postdisaster environments where conventional infrastructure may be compromised. This study presents a privacy-preserving framework for postdisaster structural health monitoring (SHM) by integrating 6G-enabled Internet of Drone Things and spatial crowdsourcing. Drones and unmanned ground vehicles collect real-time imagery of damaged infrastructure. To address privacy concerns and reduce communication overhead, we employ federated learning (FL), which enables decentralized model training on local devices without transmitting raw data. A key challenge in FL is the presence of nonindependent and identically distributed data across clients, which degrades global model performance. To mitigate this, we propose personalized conditional federated averaging (PC-FedAvg), a selective aggregation method that incorporates only client models with low validation loss into the global update. The PC-FedAvg framework is built on EfficientNetV2 and includes personalized model adaptation to enhance generalization on heterogeneous data. Experimental results on the “Concrete Crack Images for Classification” dataset demonstrate that PC-FedAvg outperforms baseline FL methods in accuracy and stability. This approach improves the effectiveness and reliability of SHM systems in real-world postdisaster scenarios by enabling timely and accurate damage assessment while preserving data privacy.
In this study, we present two configurations of piezo-actuated microspeakers, which were fabricated by combining a self-developed aerosol deposition method with the metal MEMS microfabrication process. The stainless steel used was structurally designed to enhance the displacement amplitude of the speaker, which is related to its sound pressure level. The two packaged speakers were measured using the IEC 60318-4 standard. The package around the speaker contains a printed circuit board with the dimensions in 20.0 mm × 13.0 mm × 3.0 mm. In an enclosed field test, the bimorph single-layer (BSL) configuration reached sound levels of 98.4 dB and 92.4 dB using driving voltages of 30 Vpp and 15 Vpp at 1 kHz, respectively; however, the bimorph multi-layer (BML) configuration reached higher levels of 108.2 dB and 102.2 dB under the same conditions.
Consistency relations of internal gravity waves (IGWs) describe ratios of cross-spectral quantities as functions of frequency. It has been a common practice to evaluate the measured or simulated signals (e.g., time series of velocity, density, etc.) against the consistency relations, as a way to determine whether an oceanic field of interest is comprised of IGWs. One such study is carried out in Nelson et al. (JGR Oceans, 125(5), 2020, e2019JC015974), which certifies that the ocean interior field in a numerical simulation of a region southwest of Hawaii is dominated by IGWs, through evaluating the consistency relations derived from time series at a depth of 620 m. However, we find that when the same procedure is applied at greater depths (e.g., 2362 m, 3062 m, and 4987 m), a clear deviation of the simulated signal from the classical consistency relations is observed. In this paper, we identify the reason for the unexpected deviation and show that it is a general phenomenon due to interference of low vertical modes under the reflection by the ocean bottom. We further derive a new set of formulae to characterize the consistency relations of these low modes and validate these formulae using model output.
Numerical weather prediction of sea surface temperature (SST) is crucial for regional operational forecasts. Deep learning offers an alternative approach to traditional numerical general circulation models for numerical weather prediction. In our previous work, we developed a sophisticated deep learning model known as the Attention-based Context Fusion Network (ACFN). This model integrates an attention mechanism with a convolutional neural network framework. In this study, we applied the ACFN model to the South China Sea to evaluate its performance in predicting SST. The results indicate that for a 1-day lead time, the ACFN model achieves a Mean Absolute Error of 0.215 °C and a coefficient of determination (<inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msup><mi>R</mi><mn>2</mn></msup></semantics></math></inline-formula>) of 0.972. In addition, in situ buoy data were utilized to validate the forecast results. The Mean Absolute Error for forecasts using these data increased to 0.500 °C for a 1-day lead time, with a corresponding <inline-formula><math xmlns="http://www.w3.org/1998/Math/MathML" display="inline"><semantics><msup><mi>R</mi><mn>2</mn></msup></semantics></math></inline-formula> of 0.590. Comparative analyses show that the ACFN model surpasses traditional models such as ConvLSTM and PredRNN in terms of accuracy and reliability.
To assess the effectiveness of subgrid scale (SGS) models on the prediction results of unsteady loads and turbulent fluctuation of pumpjet propulsors equipped with both front and rear stators, a pumpjet propulsor computational model with attached parts at the model scale is developed using a fully structured mesh, and large eddy simulations are conducted. The computational results of the different SGS models are compared based on five aspects: open water characteristics, turbulence parameters, incoming turbulence spectrum, vortex structure, and fluctuating pressure. Their results are also compared with the experimental values, and the correlation between the internal flow characteristics of the pumpjet propulsor and the turbulent fluctuation is analyzed. According to the results, as regards the prediction of the open water performance of the pumpjet propulsor containing both front and rear stators, the overall trend obtained by the three subgrid models is similar, and the error between the values predicted by the SL model and the experimental ones is the smallest. At the same mesh level, the turbulent fluctuating scale obtained by the SL model is larger than that obtained by the WALE and DSL models, and the turbulent time scale obtained by the DSL model has the smallest fluctuation in the circumferential direction. Among the three SGS models, the turbulent fluctuating scale of the SL model is larger than those of the WALE and DSL models. The SL model exhibits the largest energy dissipation among the three SGS models, followed by the DSL model, while that of the WALE model is the smallest. In the WALE model, the leakage vortex at the top of the blade is the longest, followed by the DSL model, while it is the shortest in the SL model. In the WALE and DSL models, the fluctuating load fluctuates more in the transition region from the middle section to the trailing edge of the blade.
Large Language Models (LLMs) have significantly advanced software engineering (SE) tasks, with prompt engineering techniques enhancing their performance in code-related areas. However, the rapid development of foundational LLMs such as the non-reasoning model GPT-4o and the reasoning model o1 raises questions about the continued effectiveness of these prompt engineering techniques. This paper presents an extensive empirical study that reevaluates various prompt engineering techniques within the context of these advanced LLMs. Focusing on three representative SE tasks, i.e., code generation, code translation, and code summarization, we assess whether prompt engineering techniques still yield improvements with advanced models, the actual effectiveness of reasoning models compared to non-reasoning models, and whether the benefits of using these advanced models justify their increased costs. Our findings reveal that prompt engineering techniques developed for earlier LLMs may provide diminished benefits or even hinder performance when applied to advanced models. In reasoning LLMs, the ability of sophisticated built-in reasoning reduces the impact of complex prompts, sometimes making simple zero-shot prompting more effective. Furthermore, while reasoning models outperform non-reasoning models in tasks requiring complex reasoning, they offer minimal advantages in tasks that do not need reasoning and may incur unnecessary costs. Based on our study, we provide practical guidance for practitioners on selecting appropriate prompt engineering techniques and foundational LLMs, considering factors such as task requirements, operational costs, and environmental impact. Our work contributes to a deeper understanding of effectively harnessing advanced LLMs in SE tasks, informing future research and application development.
Active Learning (AL) is a well-known teaching method in engineering because it allows to foster learning and critical thinking of the students by employing debate, hands-on activities, and experimentation. However, most educational results of this instructional method have been achieved in face-to-face educational settings and less has been said about how to promote AL and experimentation for online engineering education. Then, the main aim of this study was to create an AL methodology to learn electronics, physical computing (PhyC), programming, and basic robotics in engineering through hands-on activities and active experimentation in online environments. N=56 students of two engineering programs (Technology in Electronics and Industrial Engineering) participated in the methodology that was conceived using the guidelines of the Integrated Course Design Model (ICDM) and in some courses combining mobile and online learning with an Android app. The methodology gathered three main components: (1) In-home laboratories performed through low-cost hardware devices, (2) Student-created videos and blogs to evidence the development of skills, and (3) Teacher support and feedback. Data in the courses were collected through surveys, evaluation rubrics, semi-structured interviews, and students grades and were analyzed through a mixed approach. The outcomes indicate a good perception of the PhyC and programming activities by the students and suggest that these influence motivation, self-efficacy, reduction of anxiety, and improvement of academic performance in the courses. The methodology and previous results can be useful for researchers and practitioners interested in developing AL methodologies or strategies in engineering with online, mobile, or blended learning modalities.
Dhruv Bhagtani, Andrew McC. Hogg, Ryan M. Holmes
et al.
The North Atlantic Ocean circulation, fuelled by winds and surface buoyancy fluxes, carries 1.25 PettaWatts of heat poleward in the subtropics, and helps in regulating global weather and climate patterns. Here, we assess the relative impacts of changes in winds and surface heat fluxes on the Atlantic Ocean circulation and heat transport on short timescales (<10 years) and long timescales (>50 years) using ocean simulations. We decompose the circulation and heat transport into warm and cold cells (resembling a subtropical gyre and the dense overturning circulation respectively), and a mixed cell capturing waters transitioning between warm and cold regions. Warm and mixed cells transport more heat poleward as wind stress increases; however, these anomalies are compensated by reductions in the cold cell's heat transport. Warm and cold cells transport more heat poleward when we increase meridional heat flux gradients. Our findings underscore the distinct roles of winds and surface heat fluxes in controlling the Atlantic Ocean's meridional heat transport.
There are increasing evidences that the biodilution effect can significantly reduce the biomagnification of mercury (Hg) in fish. The significant antagonism of selenium (Se) -Hg may have a potential diluting effect on Hg in fish; however, there is still lack of knowledge on such effect. To reveal the Se-Hg interaction and its role in controlling the biodilution effect of Hg, we investigated levels of Hg and Se in the muscle and liver of redlip mullet from Jiaozhou Bay, China, an urbanized semi-enclosed bay highly impacted by human activities. In general, Hg levels in fish muscle were significantly negatively correlated to the levels of Se in the liver and fish size for fish with a size of < 200 mm, indicating that the antagonistic effect of Se on Hg increased with fish growth. This relationship was not significant for fish with a size of > 200 mm, possibly because the normal metabolism of Hg in muscle was hindered by homeostatic regulation or physiological activities such as gonadal development in vivo. Furthermore, the molar ratio of Se in the liver/Hg in the muscle was significantly increasing with Se/Hg in the liver, suggesting that the liver may be the key organ involved in Se-Hg antagonism. Moreover, both ratios continued to decrease with increasing fish size, implying that the antagonistic effect weakens with fish growth. These results indicate that Hg sequestration by liver may be a key mechanism of Se-Hg antagonism in fish and function as a driver for the biodilution effect of Hg, especially at a size of < 200 mm. These findings are further supported by the established linear model of Se-Hg antagonism at different developmental stages.
The failure of mooring lines in net cages could lead to the death or escape of farmed fish, which causes huge economic losses and immeasurable ecological impacts. Therefore, it is very important to ensure the safety of the cage mooring system in practical applications. This study experimentally and numerically investigates the hydrodynamic behaviors of the mooring line failure on a net cage in irregular waves and currents. For the model test, a 1:25 scaled net cage model with eight-point mooring is installed in a wave tank. The two load cells in its upstream mooring lines and a gyroscope on its floating collar are used to measure the mooring force and the rotational motions, respectively. A cutting device equipped with a pneumatic cylinder and a blade is used to cut the line. A self-developed numerical model is specifically established for the model test for cross-validation. Both experimental and numerical results are analyzed and compared in the time and frequency domains. The results show that the mooring load in the remaining line significantly increases as one of the upstream mooring lines is disconnected. Meanwhile, a significant yaw rotation of the floating collar is observed. The results indicate that the maximum tension, drift displacement, and rotational angles significantly increase as the current velocity increases.
Science, General. Including nature conservation, geographical distribution
Katharina Gallmeier, J. Xavier Prochaska, Peter C. Cornillon
et al.
We extracted ~2.8M nearly cloud-free 144x144 km^2 cutout images from the 2012-2020 Level-2 VIIRS Sea Surface Temperature (SST) dataset to quantitatively compare with MIT ocean general circulation model outputs, specifically the one year LCC4320 1/48 deg global-ocean simulation starting on November 17, 2011, matched in geography and day-of-year to VIIRS observations. We analyzed these cutouts using an unsupervised probabilistic autoencoder (PAE) to learn the SST structure on ~10-to-80 km scales (submesoscale-to-mesoscale). A key finding is that, in general, the LLC4320 simulation accurately reproduces the observed SST patterns, both globally and regionally. Global structure distribution medians match within 2 sigma for 65% of the ocean, despite a modest, latitude-dependent offset. Regionally, model outputs mimic mesoscale SST pattern variations in VIIRS data revealed by PAE, including subtle features influenced by bathymetry variations. There are however some areas showing significant differences in the distribution of SST patterns: (1) near western boundary currents' separation from the continental margin, (2) in the ACC, particularly in the eastern half of the Indian Ocean, and (3) in an equatorial band equatorward of 15 deg. The discrepancy in (1) results from premature separation of simulated western boundary currents. In (2), the Southern Indian Ocean, the model output predicts more structure than observed, possibly due to mixed layer misrepresentation or energy dissipation and stirring inaccuracies in the simulation. The differences in (3), the equatorial band, may also stem from model errors, potentially arising from the simulation's shortness or insufficient high-frequency/wavenumber atmospheric forcing. While the exact causes of these model-data differences remain uncertain, such comparisons are expected to guide future developments in high-resolution global-ocean simulations.
Quantum software engineering (QSE) is receiving increasing attention, as evidenced by increasing publications on topics, e.g., quantum software modeling, testing, and debugging. However, in the literature, quantum software requirements engineering (QSRE) is still a software engineering area that is relatively less investigated. To this end, in this paper, we provide an initial set of thoughts about how requirements engineering for quantum software might differ from that for classical software after making an effort to map classical requirements classifications (e.g., functional and extra-functional requirements) into the context of quantum software. Moreover, we provide discussions on various aspects of QSRE that deserve attention from the quantum software engineering community.
The spatiotemporal variation in tropical air-sea interaction is investigated by applying a simple model that considers the fundamental dynamics in tropical oceans. The model decomposes sea surface temperature anomaly (SSTA) variation into a series of spatial modes that oscillates with their natural frequencies. The results suggest that the first mode associates with the dipole-like SSTA variation between the western and the eastern coast, such as EP El Niño, the Atlantic Niño, and IOD; whereas the second mode associates with the tripole-like SSTA pattern among the central and eastern, western coast, such as CP El Niño and minor SSTA variations in the tropical Atlantic and Indian Ocean. Each mode oscillates with its natural frequency that depends on the strength of air-sea coupling and the basin size. The model provides a systematic framework for the comprehensive understanding of the complex air-sea interaction in tropical oceans.
Alexander E. I. Brownlee, James Callan, Karine Even-Mendoza
et al.
Large language models (LLMs) have been successfully applied to software engineering tasks, including program repair. However, their application in search-based techniques such as Genetic Improvement (GI) is still largely unexplored. In this paper, we evaluate the use of LLMs as mutation operators for GI to improve the search process. We expand the Gin Java GI toolkit to call OpenAI's API to generate edits for the JCodec tool. We randomly sample the space of edits using 5 different edit types. We find that the number of patches passing unit tests is up to 75% higher with LLM-based edits than with standard Insert edits. Further, we observe that the patches found with LLMs are generally less diverse compared to standard edits. We ran GI with local search to find runtime improvements. Although many improving patches are found by LLM-enhanced GI, the best improving patch was found by standard GI.