Melt ponds are puddles of water on Arctic summer sea ice that play an important role in the Arctic climate system, mainly through their influence on the surface energy balance. The albedo of polar oceans is primarily determined by the area fractions of sea ice and darker open ocean. However, the presence of melt ponds also significantly reduces the reflective properties of sea ice. This allows more solar energy to be absorbed, accelerating regional ice melt in a phenomenon known as ice-albedo feedback. Despite their importance, there is a lack of observational datasets that explicitly parameterize melt ponds next to sea ice and open water, and the existing data are mainly based on optical imagery. In this study, we present a new helicopter-borne dataset based on (TIR) and employ a deep learning-based semantic segmentation approach to classify surface types (melt ponds, sea ice, and open water). We evaluate the performance of several model architectures: U-Net, UNet++, PSPNet, DeepLabV3+, SwinTransformer, SAM2-UNet, and AutoSAM. Among these models, UNet++ pre-trained on the AID dataset demonstrates the highest performance, achieving a melt pond (IoU) of 0.486, a sea ice IoU of 0.764, and an open water IoU of 0.660. Furthermore, we apply the most effective model to the datasets of five helicopter flights, resulting in (MPF) estimates with a bias of approximately 11%, which is higher than for melt pond classifications based on visible observations.

Visual simultaneous localization and mapping (SLAM) systems are extensively explored due to their low cost. To tackle more challenging environments, multi-camera SLAM systems with a wider field of view (FoV) have been proposed. However, most of the existing multi-camera SLAM systems still rely on hand-crafted features. Given the remarkable advancements of deep learning techniques in the field of feature extraction, the integration of convolutional neural networks into arbitrary multi-camera SLAM systems represents a research topic worthy of in-depth exploration. In this paper, a Deep learning-based Arbitrary Multi-Camera SLAM system (DAMC-SLAM) is proposed. Specifically, this system leverages convolutional neural networks (CNNs) to perform online extraction of local features and global descriptors, while generating corresponding masks to filter out dynamic objects in the scene. Furthermore, DAMC-SLAM is designed with strong adaptability, as it can be flexibly compatible with an arbitrary number of cameras, regardless of whether there exist overlapping FoVs between the deployed cameras. Six self-collected sequences under various environmental conditions, together with the EuRoC MAV benchmark, were used to evaluate the performance of DAMC-SLAM. Experimental results demonstrate that our system possesses remarkable robustness and superior performance. Compared with other methods, it achieves consistent and successful tracking performance across all self-collected sequences, while concurrently yielding the minimum absolute trajectory error and scale error over these sequences. It also demonstrates competitive performance on the EuRoC MAV benchmark, achieving the lowest average ATE among the evaluated methods.

Species trait data are widely used in ecology and evolution to approximate ecological niche. Despite this, the data housed in species trait databases are often patchy or biased toward charismatic species and easy-to-study traits. Filling data gaps can be achieved by determining the most likely values via data imputation. We conducted a systematic literature review to identify data imputation methods for species trait data in ecological and evolutionary studies, as well as in data science studies, and compared the two domains to identify whether alternative methods might be suitable for adoption in ecology and evolution. We focus on novel and state-of-the-art deep learning (DL) imputation methods. We use evidence maps to identify missing data contexts where DL alternatives have been shown to be effective in data science. Differences in preferred methods were identified between the two domains. DL methods were rarely used in ecological and evolutionary studies, whereas they were frequently employed in data science. Several DL methods that have not been adopted in ecology and evolution demonstrated state-of-the-art performance when imputing data in contexts that are similar to those faced by ecologists wishing to impute species trait data. We identified a clear gap in the ecological and evolutionary literature in relation to the uptake and use of imputation methods based on DL. We then identify autoencoders, transformer-based architectures, Bayesian neural networks, and feed-forward neural networks as primary architectures worth further investigation for their capacity to impute missing species trait data and potential for multiple imputation.

With the ongoing effects of global warming, Antarctic supraglacial meltwater plays a critical role in ice-sheet instability. Spaceborne remote sensing provides a viable solution to detect surface melt in these vast and remote regions. Innovative deep learning image segmentation methods have the potential to improve supraglacial lake (SGL) mapping by adding spatial awareness to the mapping process, as currently established methods only consider the spectral variety of multispectral satellite data (Sentinel-1 and Landsat-8) on a per-pixel basis. In this study, we adapt a U-net image segmentation method to detect SGLs in Sentinel-2 data and quantify its performance in comparison to established mapping approaches in Antarctica. At a 90% probability threshold, the proposed workflow achieves a producer’s accuracy, user’s accuracy, and an F1 score of 0.915, 0.912, and 0.913, respectively. We compare the classification results to those of an established Random Forest classifier, and find that the U-net achieves better results in 13 of the 16 validation scenes. In a relative comparison with five established SGL products, the U-net approach maps, on average, the largest lake area, although there is disagreement between the datasets due to different data sources, cloud cover, and misclassification of some clouds as lakes. The main advantages of the developed approach are its capabilities to map lakes under thin clouds and ice, resulting in more continuous lake areas than conventional approaches, and a reduced number of input features required, which will be advantageous in future large-scale SGL mapping efforts.

While deep learning (DL) methods have expanded the approach to solving state estimation problems in data assimilation (DA), a robust approach requires extending them for uncertainty quantification. In this work, we propose a method to represent location-based uncertainty through random, spatiotemporally coherent displacements in physical space, which are used to sample states in the vicinity of a given reference state. We integrate this strategy into an end-to-end variational DA framework, termed 4DVarNet-LU, and compare its performance with a UNet baseline, demonstrating improved reconstruction skill through the explicit use of prior state information. Trained deterministically in output and stochastically in input, we demonstrate that at inference, we see improvement in skill when using information from initial conditions. Using observation system simulation experiments based on a single-layer quasi-geostrophic model with sparse pseudo-nadir altimetry observations, we evaluate reconstruction mean and ensemble skill. Results show that 4DVarNet-LU consistently outperforms the UNet baseline in terms of root mean square error (RMSE) and continuous ranked probability score (CRPS), with increased robustness to initial condition bias and improved temporal stability. While the method does not propose to account for all different uncertainty sources in DA, it provides a structured way for modeling and training DL models for dealing with location-based uncertainty. Future work may extend this approach to account for additional sources of uncertainty and end-to-end ensemble generation.

Sentimental analysis (SA) of movie reviews has become an essential means of supporting audiences, filmmakers, and investors to facilitate data-driven marketing, support audience engagement, and maximize audience development returns. Nevertheless, current SA models are yet to overcome potentially daunting problems such as noisy data, ambiguity of language, limited handling of negations, and a loss of rich semantic information during mirror preprocessing that ultimately decreases classification performance. To overcome this hurdle, a unique two-stage strategy entitled Hierarchical Attention-Guided Sentiment Network (HASN) is proposed that enables reliable and interpretable sentiment predictions. The first stage includes a clearly defined preprocessing pipeline of the IMDb dataset that features text cleaning, lemmatizing, removing stop words, negation, TF-IDF filtering of low-frequency terms, and lastly, Word2Vec embeddings, which create semantically consistent and contextually rich representations of potential inputs. The second stage integrates a tri-stream attention recurrent extractor (TARE), which reconstructs and merges previously learned outputs with the learned syntactic and semantic contexts with the benefits of stacked Bi-LSTMs, combines learned syntactic and semantic contexts with multi-head attention to identify the tokens that drive sentiment, and ends with a residual-normalized GRU (ResNorm-GRU), which sequentially refines the sequential process of intake. The attentive conditional random field (Attn-CRF) serves as a classifier by boosting coherence in label predictions through attention weighting with modeling of the sequential dependencies. The experiments with the IMDb dataset demonstrated that the best results detected were the following: 98.09, 98.47, 97.70, and 98.08%. In summary, the HASN framework is a precise, scalable, and interpretable model for analyzing sentiments in movie reviews.

Maintaining epidermal integrity is essential for plant growth and survival, yet quantitative methods to assess defects in cell adhesion, cell integrity and cuticle function remain limited. Here, we present RRQuant, a standardised, high-throughput workflow for quantifying epidermal integrity defects in seedlings using Ruthenium Red staining. RRQuant integrates deep learning–based segmentation with automated image analysis, staining and morphological quantification, and statistical evaluation. The workflow is accompanied by a detailed user guide, and an interactive R Shiny application for data visualisation and quality control. This pipeline converts qualitative staining patterns into reproducible quantitative measurements, enabling large-scale phenotyping and comparative studies across genotypes and treatments. We demonstrate the usefulness of RRQuant by quantitatively comparing adhesion defective phenotypes and discuss its potential for adaptation to other organs, staining methods, and pigment-based assays. By providing an open and extensible platform, RRQuant bridges the gap between traditional qualitative assessments and modern quantitative phenotyping in plant biology.

Human parasitic gastrointestinal infections affect over 3.5 billion people worldwide, yet their diagnostic management remains heavily dependent on conventional microscopy. While artificial intelligence (AI) has emerged as a valuable tool in digital parasitology, its application has progressed unevenly across different parasite classes. This systematic review evaluates AI-based diagnostic methods for human gastrointestinal parasites, systematically comparing performance between helminth detection and protozoan identification. From 4323 identified records, 77 studies met our inclusion criteria after rigorous application of exclusion criteria. Of these, 42 studies (54.5%) focused on helminth detection, 28 studies (36.4%) focused on gastrointestinal protozoa and 7 studies (9.1%) evaluated algorithms applicable to both parasite classes. Convolutional neural networks dominated the field (66.2% of studies), with You Only Look Once variants representing the most common object detection architecture (27.3%). For helminth detection, reported sensitivities ranged from 85% to 98% in externally validated studies, with several systems demonstrating superior sensitivity compared to expert microscopists. For protozoan detection, although internal validation frequently reported accuracies exceeding 90%, only 17.9% of studies performed external validation, which revealed substantial performance degradation of 15–30 percentage points. AI demonstrates promising performance in controlled settings, yet it remains limited by a lack of external validation. Notably, 71.4% of the datasets were sourced from a single institution, restricting the scope of generalizability claims. Closing this diagnostic gap requires coordinated efforts to establish standardized imaging protocols, create diverse open-access repositories and adapt advanced machine learning techniques from adjacent fields such as digital pathology.

Connecting multiple destinations with the shortest overall path in an obstacle-rich indoor environment is a classic problem in robotic path planning. Established solutions, such as A*, ARA*, and RRT*, execute faster when given smaller input maps that exclude extraneous paths. However, removing such paths without losing the ground truth solution is challenging because runtime-efficient approximations can yield detours. This work proposes a deep learning-based search-space reduction pipeline (SRP) that integrates a module called MazeNet to solve indoor navigation tasks with fast runtimes while maintaining accuracy. We transform the indoor environment into a graph representation, reducing the planning task into a graph-based optimization problem known as the Obstacle-Avoiding Rectilinear Steiner Minimum Tree. MazeNet takes compressed images derived from graph abstractions as inputs and uses a recurrent convolutional neural network trained to predict the shortest feasible path connecting all destinations. Within SRP, MazeNet’s output restricts the search space for classical planners. MazeNet solves more complex instances than those seen during training via recurrence with an integrated termination condition. We evaluate MazeNet’s runtime and path-length performance using a variety of planning methods against both exact and approximate graph algorithms in simulations and, after fine-tuning, in real-life floor plans. MazeNet solves all these specific test cases with improved runtimes and no increase in path length. Physics simulation further supports SRP’s feasibility across the tested floor plans, with no collisions recorded during navigation tasks.

Structural health monitoring plays a crucial role in ensuring the safety, reliability, and longevity of bridge infrastructures through early damage detection. Although recent advances in deep learning-based models have enabled automated event detection, their performance is often limited by data scarcity, environmental noise, and class imbalance. To address these challenges, this study introduces a customized generative adversarial network (GAN) model, STFTSynth, designed particularly for generating short-time Fourier transform spectrograms derived from acoustic-event signals. In contrast to augmentation techniques such as MixUp, GANs can synthesize spectrograms that visually and statistically resemble real event representations, providing a basis for representation-level dataset enrichment. The proposed model integrates dense residual blocks for spatial consistency with bidirectional gated recurrent units for temporal dependency modeling. Model performance is evaluated against three baseline generative models using qualitative inspection and quantitative metrics, including structural similarity index measure, peak signal-to-noise ratio, and Fréchet inception distance. Results show that STFTSynth outperforms baseline models, producing high-resolution, temporally consistent spectrograms that align closely with real-world data. These findings highlight the potential of GAN-based spectrogram synthesis for representation-level enrichment of rare acoustic-event datasets in bridge monitoring, particularly when real examples such as prestressing wire breakage are limited.

In insurance underwriting, applicants may deliberately misrepresent certain risk factors to secure lower premiums. However, from the insurer’s perspective, verifying the true status of these risk factors is both time-consuming and resource-intensive. To address this challenge, we propose a novel deep learning framework for claim severity modeling that explicitly accounts for potential misrepresentation of a binary risk factor. To mitigate the identifiability issues arising from the unobservable nature of misrepresentation in neural networks, we introduce a regularization term based on Kullback–Leibler (KL) divergence. This term serves as a conservative anchor by penalizing deviations from a baseline assumption of honesty, unless such deviations are strongly supported by empirical evidence. Additionally, we introduce an innovative regularization term, the $L_1$ group minimax concave penalty ($L_1$-gMCP), within the neural network’s loss function. This term facilitates effective variable selection in high-dimensional settings. Through comprehensive simulation studies, we demonstrate that our method maintains robust performance regardless of whether misrepresentation is present in the data. The proposed model excels in both prediction accuracy and the identification of relevant risk factors. We further validate our approach using real-world data from the 2014 Medical Expenditure Panel Survey, treating reported insurance status as a potentially misrepresented variable.

Machine learning-based weather forecasting models now surpass state-of-the-art numerical weather prediction systems, but training and operating these models at high spatial resolution remains computationally expensive. We present a modular framework that decouples forecasting from spatial resolution by applying learned generative super-resolution as a post-processing step to coarse-resolution forecast trajectories. We formulate super-resolution as a stochastic inverse problem, using a residual formulation to preserve large-scale structure while reconstructing unresolved variability. The model is trained with flow matching exclusively on reanalysis data and is applied to global medium-range forecasts. We evaluate (i) design consistency by re-coarsening super-resolved forecasts and comparing them to the original coarse trajectories and (ii) high-resolution forecast quality using standard ensemble verification metrics and spectral diagnostics. Results show that super-resolution preserves large-scale structure and variance after re-coarsening, introduces physically consistent small-scale variability, and achieves competitive probabilistic forecast skill at 0.25° resolution relative to an operational ensemble baseline while requiring only a modest additional training cost compared with end-to-end high-resolution forecasting.

If crevasse fields deliver meltwater to the bed of the Greenland ice sheet, it would affect seasonal ice flow speeds and total mass balance. The best current automated tool to map crevasse fields extends only a few dozen kilometers inland. To address this gap, we develop MimiNet, a neural-network-based tool that identifies surface crevasse fields. We train MimiNet on Sentinel-1 scenes across a 629 km2 area in Pâkitsoq, central-western Greenland, and use it to locate crevasse fields annually over 2015–24. We find that the crevassed area varied from a minimum of 106 ± 5 km2 in 2019 to a maximum of 144 ± 6 km2 in 2016, with no overall trend over the 10 year study period. We find some evidence that seasonal ice velocity anomalies in crevasse fields are higher than those in moulin-drained areas in the late melt season. This may suggest that the subglacial drainage system under crevasse fields remains inefficient all summer, and thus that at least some Pâkitsoq crevasse fields deliver meltwater to the bed. Interannual variability in ice dynamics may drive the observed variability in crevassed areas; we expect crevasse extent to become more variable in time as ice flow speeds and their variations, amplify under climate change.

In recent years, deep learning has transformed data processing, justifying its use in the current study. These models consist of multiple layers that calibrate weights to optimize accuracy. By mimicking neural processes, these networks facilitate high-performance predictive modeling. This study focused on the application of convolutional neural networks for forecasting Sargassum seaweed arrivals in Guadeloupe, using varied climatic data such as winds, surface currents, satellite images, and observations of seaweed arrivals on the Guadeloupe coast. We tested two different approaches: classification models for forecasting up to 14 days and zone regression models for forecasting up to 14 days, also indicating a probability of algae arrival. The results showed that the classification models achieved a mean accuracy of 94.75%, a mean specificity of 97.42%, and a mean sensitivity of 98.78%. Zone regression models achieved a mean accuracy of around 90%, a mean specificity of around 97%, and a mean sensitivity of around 85%, with a mean MAE of 0.13 and a mean RMSE of 0.18. These models perform better than the decision tree proposed in the state of the art for the same problem.

This paper presents a deep learning-based approach to automatically classify the rust level of screws using ResNet-18 and MobileNetV3 convolutional neural networks. A controlled salt-spray chamber was used to simulate corrosion on metal screws over 0h, 48h, 96h, and 168h of exposure. Images were processed with a circle-detection algorithm to extract individual screws, followed by data augmentation and training. The final models achieved a classification accuracy greater than 94% on the validation set.

Modern aviation supports an ever-broader range of civil and military missions, and the airframes designed for these missions must satisfy stringent safety and performance requirements. The takeoff and landing phases are the most accident-prone portions of a flight despite representing only a short interval of the total block time, which makes the accurate prediction of takeoff speed a safety-relevant problem. A previous machine learning study addressed the takeoff-speed prediction problem of the Boeing 737-300 with classical regressors using pressure altitude, outside air temperature, gross weight and flap angle as the predictors. In the present work, the same regression problem is revisited under the deep learning paradigm. Four neural architectures are trained on an identical pre-processing pipeline and train-validation partition, namely a multilayer perceptron, a one-dimensional convolutional network, a long short-term memory network and a wide-and-deep architecture incorporating multi-head self-attention. Among the four candidates, the long short-term memory network attains the lowest root mean square error and mean square error on the unseen test file and is subsequently subjected to Bayesian hyperparameter optimisation through the Keras Tuner library. The predicted and the measured takeoff speeds are reported side by side for the first time in the deep learning literature for this airframe, and the simulation results indicate that the developed networks constitute an effective alternative tool for takeoff-speed prediction.