72. Open access datasets for implementing computer vision in PLF

Editorial and News

Here we present data from two studies, both of which had the purpose of investigating the potential of using electroencephalograms measured from the ear (’ear-EEG’) for sleep monitoring in a home environment. In total, 320 nights were recorded. All nights were recorded with ear-EEG, and some were also recorded using scalp-EEG and/or wristworn actigraphy. All subjects were recorded multiple times. To our knowledge, this is the most extensive open access data set available for mobile EEG development, and possibly also the best open access dataset for studying repeated sleep monitoring on individuals. We describe the details of each data set, including data quality measures, and compare the sleep scoring performance to a previously published dataset.

Aim: In addition to affecting the individual sociologically and psychologically, heart disease also poses important problems in health systems. Evaluation of heart disease performances has gained great importance in terms of machine learning method. In the study, performances were compared with the machine learning method for risk methods that classify heart illness. Materials and Methods: The categorization process Throughout the research made use of the "Heart Disease Dataset," an open access dataset. F1-score, sensitivity, selectivity, accuracy, balanced accuracy, negative and positive predictive values were used to assess the performance of the categorisation model using the machine learning approach. Random forest method, one of the variable selection methods, was used. Results: According to the relational classification model's classification findings for heart disease, the accuracy, balanced accuracy, sensitivity, selectivity, positive predictive value, negative predictive value, and F1-score values were observed to be 0.997, 0.997, 0.995, 1, 1, and 0.995, respectively. Conclusion: The relational classification model proposed in the analysis obtained in the web-based open access dataset yielded distinctively successful results in classifying heart disease according to performance criteria.

Global water systems are under increasing pressure from climate change, hydrological extremes, and competing demands on limited freshwater resources. Robust, interoperable data and data sharing infrastructures are essential to advance scientific understanding, support operational forecasting, and inform effective water management and climate adaptation strategies. The Reference Observatory of Basins for INternational hydrological climate change detection (ROBIN) and the Global Hydrological Status and Outlook System (HydroSOS) represent complementary international efforts to unify and leverage hydrological data for science and service delivery. ROBIN, coordinated by the UK Centre for Ecology & Hydrology, addresses critical gaps in global streamflow observations by integrating long-term, near-natural catchment data to form a global Reference Hydrometric Network. This open-access dataset, comprising >3000 of quality-controlled streamflow records from near-natural catchments spanning diverse climates and geographies, is shared under common standards and protocols to enable global-scale climate change trend detection, model evaluation, and large-scale hydrological research. The initiative also fosters long-term collaboration among researchers and institutions, promoting shared infrastructure, code libraries, and harmonised metadata to support international scientific agendas. ROBIN is aligning its data-sharing processes with the GRDC, ensuring interoperability and complementarity between datasets, and strengthening the long-term international hydrological data legacy. Through collaboration with the FRIEND-Water and EUROFRIEND networks, ROBIN is increasing its emphasis on social hydrology, enabling analyses that better capture human–water interactions, vulnerability, and adaptation alongside climate-driven hydrological change. The World Meteorological Organization’s HydroSOS builds operational capacity for standardised hydrological status assessments and sub-seasonal to seasonal outlooks across spatial scales. HydroSOS enhances national and regional water information systems by linking local observations, forecasts, and global services to produce consistent products for water resources management, disaster risk reduction, and climate resilience. Developed and supported by the UKCEH, the HydroSOS portal provides open access to these standardised “change from normal” assessments, enabling transparent comparison of hydrological conditions across countries and regions. It emphasises the value of consistent and standardised hydrological information among National Meteorological and Hydrological Services, contributing to united global water information frameworks and decision support tools. Together, ROBIN and HydroSOS exemplify synergistic efforts to overcome current and historical fragmentation in hydrological data and services. By promoting open data practices, harmonised standards, and shared technical capacity, these initiatives enable a more integrated global hydrological data ecosystem that supports science, policy, and operational services in the face of evolving environmental challenges and a need to improve resilience to extreme hydrological events both now and in the future.

No one refutes the importance of datasets in the development of any new approach. Despite their importance, open access datasets in computer vision remain insufficient for some applications. This paper introduces a FallFree, new and rich dataset that can be used for the evaluation/development of computer vision-based applications pertinent to people who use a cane as a mobility aid, e.g., fall detection, activity recognition. In particular, the FallFree dataset includes video streams captured with Kinect which offers a wide range of visual information. It is organized hierarchically, in terms of scenarios each of which is structured in terms of its features. The current FallFree dataset version covers all fall scenarios of the cane users along with various non-fall scenarios grouped into one set. Each scenario is represented through a rich set of features that can be extracted from Kinect. To widen its usability, the dataset was constructed while accounting for existing datasets' organization, size, scope, streams, types and hypotheses.

In recent years, the development of computer vision technology has led to significant implementations of non-contact structural identification. This study investigates the performance offered by the Phase-Based Motion Magnification (PBMM) algorithm, which employs video acquisitions to estimate the displacements of target pixels and amplify vibrations occurring within a desired frequency band. Using low-cost acquisition setups, this technique can potentially replace the pointwise measurements provided by traditional contact sensors. The main novelty of this experimental research is the validation of PBMM-based experimental modal analyses on multi-storey frame structures with different stiffnesses, considering six structural layouts with different configurations of diagonal bracings. The PBMM results, both in terms of time series and identified modal parameters, are validated against benchmarks provided by an array of physically attached accelerometers. In addition, the influence of pixel intensity on estimates’ accuracy is investigated. Although the PBMM method shows limitations due to the low frame rates of the commercial cameras employed, along with an increase in the signal-to-noise ratio in correspondence of bracing nodes, this method turned out to be effective in modal identification for structures with modest variations in stiffness in terms of height. Moreover, the algorithm exhibits modest sensitivity to pixel intensity. An open access dataset containing video and sensor data recorded during the experiments, is available to support further research at the following https://doi.org/10.5281/zenodo.10412857.

Supervised machine learning generally requires pre-labelled data. Although there are several open access and pre-annotated datasets available for training machine learning algorithms, most contain a limited number of object classes, which may not be suitable for specific tasks. As previously available pre-annotated data is not usually sufficient for custom models, most of the real world applications require collecting and preparing training data. There is an obvious trade-off between annotation quality and quantity. Time and resources can be allocated for ensuring superior data quality or for increasing the quantity of the annotated data. We test the degree of the detrimental effect caused by the annotation errors. We conclude that while the results deteriorate if annotations are erroneous; the effect - at least while using relatively homogeneous sequential video data - is limited. The benefits from the increased annotated data set size (created by using imperfect auto-annotation methods) outweighs the deterioration caused by annotated data.

Plant traits database for climate change adaptation and mitigation in Northwest Portugal

Objective: This study aimed to compare the classification performance of acute inflammation by applying the RBF ANN model on an open-access acute inflammation data set and determining the risk factors that may be associated with acute inflammation markers. Material and Methods: In the study, Nephritis of renal pelvis origin was classified using the open access “Acute Inflammation” data set RBF ANN model, and risk factors that could be associated were revealed. The success of RBF ANN is presented by different performance metrics. Results: The success of classifying Nephritis of renal pelvis origin with the RBF ANN model has been demonstrated to be excellent (AUC = 1, Accuracy = 100%). In addition, the RBF ANN model revealed that the most important variable among the risk factors that may be associated with Nephritis of renal pelvis origin is “temperature of patient”. Conclusion: As a result, the obtained findings show that the RBF ANN model provides very successful predictions in the classification of Nephritis of renal pelvis origin. Also, it has been shown that the importance values of factors associated with Nephritis of renal pelvis origin are estimated with the RBF classification model and can be used safely in preventive medicine applications.

The human body is an outstandingly complex machine including around 1000 muscles and joints acting synergistically. Yet, the coordination of the enormous amount of degrees of freedom needed for movement is mastered by our one brain and spinal cord. The idea that some synergistic neural components of movement exist was already suggested at the beginning of the 20th century. Since then, it has been widely accepted that the central nervous system might simplify the production of movement by avoiding the control of each muscle individually. Instead, it might be controlling muscles in common patterns that have been called muscle synergies. Only with the advent of modern computational methods and hardware it has been possible to numerically extract synergies from electromyography (EMG) signals. However, typical experimental setups do not include a big number of individuals, with common sample sizes of 5 to 20 participants. With this study, we make publicly available a set of EMG activities recorded during treadmill running from the right lower limb of 135 healthy and young adults (78 males and 57 females). Moreover, we include in this open access data set the code used to extract synergies from EMG data using non-negative matrix factorization (NMF) and the relative outcomes. Muscle synergies, containing the time-invariant muscle weightings (motor modules) and the time-dependent activation coefficients (motor primitives), were extracted from 13 ipsilateral EMG activities using NMF. Four synergies were enough to describe as many gait cycle phases during running: weight acceptance, propulsion, early swing, and late swing. We foresee many possible applications of our data that we can summarize in three key points. First, it can be a prime source for broadening the representation of human motor control due to the big sample size. Second, it could serve as a benchmark for scientists from multiple disciplines such as musculoskeletal modeling, robotics, clinical neuroscience, sport science, etc. Third, the data set could be used both to train students or to support established scientists in the perfection of current muscle synergies extraction methods. All the data is available at Zenodo (doi: 10.5281/zenodo.1254380).

Artificial intelligence (AI) is being increasingly applied in the geosciences, particularly in fields like mineralogy, where it supports tasks such as mineral classification, automated thin‐section image analysis, or mineral exploration targeting. Such tasks require large structured and standardized data sets, which are currently not available. We build two databases to fill this gap: (i) MRMinerals contains a list of the 400 most common and geologically significant minerals, including major rock‐forming minerals, key accessory minerals, and economically important ore minerals with machine‐readable formulas as the key feature. (ii) MineralTD contains a large training data set with 10,000+ compositions for each of the 400 minerals in MRMinerals. MineralTD is split into two subdatasets: MineralTDMeasured and MineralTDSynthetic . MineralTDMeasured contains approximately 140,000 mineral compositions from the open‐access geochemical databases and repositories GEOROC, Pangaea, PetDB, RRUFF, and ESMD. MineralTDSynthetic contains synthetic mineral compositions, generated using machine‐readable formulas from MRMinerals, with at least 10,000 compositions per mineral. MineralTD is annotated with metadata, such as mineral frequency, rock classification, data source, and methods used to provide a full understanding of the individual data set. The MRMinerals and MineralTD are ready‐to‐use open access data sets that enable scalable, data‐driven research in mineralogy, e.g., ML applications.

Brain-computer interfaces (BCIs) recognize specific features of a person's brain signal relating to his/her intent, and output a control command that controls the outside devices or computers. BCI systems facilitate the lives of patients who cannot move any muscles but have no cognitive disorder. The high dimensions of features represent a research challenge. In recent years, especially nature inspired heuristic optimization algorithms became popular in order to eliminate unnecessary features. This paper addresses a crucial factor for effective classification of motor imaginary based EEG signals that are an optimal selection of relevant EEG features using firefly algorithm. Firefly algorithm (FA) works on the principle of directing the less shiny than the light intensity emitted by fireflies in nature towards the bright. The algorithm can adaptively select the best subset of features and improve classification accuracy. In this study, following extracted Katz Fractal Dimension based features, effective feature(s) were selected by FA. The proposed method successfully applied on open access dataset which was collected from 29 subjects. We obtained an average 76.14% classification accuracy (CA) using k-nearest neighbor classifier. This is 4.4% higher than the CA calculated by using all features. These results proved that used method is robust for this dataset.

Comparison of total emitted solar-induced chlorophyll fluorescence (SIF) and top-of-canopy (TOC) SIF in estimating photosynthesis

Deep ensemble learning based probabilistic load forecasting in smart grids

With the exponential growth of high-quality fake images in social networks and media, it is necessary to develop recognition algorithms for this type of content. One of the most common types of image and video editing consists of duplicating areas of the image, known as the copy-move technique. Traditional image processing approaches manually look for patterns related to the duplicated content, limiting their use in mass data classification. In contrast, approaches based on deep learning have shown better performance and promising results, but they present generalization problems with a high dependence on training data and the need for appropriate selection of hyperparameters. To overcome this, we propose two approaches that use deep learning, a model by a custom architecture and a model by transfer learning. In each case, the impact of the depth of the network is analyzed in terms of precision (P), recall (R) and F1 score. Additionally, the problem of generalization is addressed with images from eight different open access datasets. Finally, the models are compared in terms of evaluation metrics, and training and inference times. The model by transfer learning of VGG-16 achieves metrics about 10% higher than the model by a custom architecture, however, it requires approximately twice as much inference time as the latter.