Accurate classification of infrasound signals is an important research topic in the fields of nuclear explosion monitoring and natural disaster early warning. Current studies often rely on empirical selection of time-frequency analysis methods to characterize denoised signals, lacking a systematic evaluation of different methods under noisy conditions within a unified framework. To address this issue, this study takes six types of non-denoised infrasound signals—including nuclear tests, chemical explosions, and lightning—as research objects, and systematically compares the comprehensive performance of Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Hilbert–Huang Transform (HHT) in a convolutional neural network (CNN) classification model. Experimental results show that STFT achieves the best overall performance under the configuration of a Hanning window with a length of 1 s and NFFT = 1,024, with a classification accuracy of 98.35%, significantly outperforming CWT and HHT. Meanwhile, STFT requires the shortest computation time (0.85 s) and exhibits the fastest model convergence. Further analysis reveals that the inherent smoothing and consistency provided by the fixed window function of STFT contribute to stronger robustness and feature stability in noisy environments. This study simplifies the classification pipeline, avoids information loss and computational overhead caused by denoising preprocessing, emphasizes the importance of selecting time-frequency methods under noisy conditions, and provides a more efficient and reliable engineering solution for real-time infrasound monitoring systems, demonstrating strong practical application value.

Pump station units under prolonged high-load operation are prone to mechanical faults that threaten the safe and stable operation of water diversion projects. Existing diagnostic methods often face challenges in adaptive parameter optimization of variational mode decomposition (VMD), modal aliasing, and insufficient spatiotemporal feature representation. To address these issues, this study proposes an intelligent fault diagnosis framework based on an improved VMD–convolutional neural network (CNN)–long short-term memory (LSTM)-coupled model. The main contributions are as follows. 1) A dedicated parameter optimization strategy is proposed by enhancing the sparrow search algorithm (SSA) with an Osprey-inspired exploration mechanism and a Cauchy mutation operator (resulting in OCSSA). This approach adaptively optimizes VMD parameters, thus overcoming the limitations of manual tuning and local optima. 2) The optimal intrinsic mode function (IMF) is selected based on envelope entropy to effectively mitigate modal aliasing and noise interference. 3) A CNN–LSTM hybrid architecture is constructed to achieve joint spatiotemporal modeling—CNN extracts local spatial features, while LSTM captures temporal dependencies—addressing the shortcomings of single models in comprehensive feature representation. Fault classification is completed via a fully connected layer and a softmax function. Experimental results show that under 5 dB low signal-to-noise ratio (SNR) conditions, the proposed model achieves 80.95% diagnostic accuracy for typical faults such as rotor misalignment and rubbing—a 12.72 percentage point improvement over the baseline CNN–LSTM model—while maintaining competitive training efficiency. Under 20 dB SNR, the accuracy further reaches 97.50%. The model significantly reduces misdiagnosis rates for complex coupled faults, demonstrating superior robustness and engineering applicability. This integrated framework offers a reliable and deployable solution for the intelligent maintenance of pump station units.

Gliomas are among the most severe types of brain tumors and can be life-threatening without early detection. Accurate and timely segmentation of brain tumors from MRI scans is crucial for effective treatment planning; however, it remains challenging due to significant variation in tumor shape, size, and location. This paper proposes a 2D Patch-wise Deep Residual U-Net with 102 convolutional layers for automatic tumor segmentation. The approach divides MRI scans into uniform, non-overlapping patches to achieve precise localization and better preserve local features. Residual blocks with identity mapping help mitigate vanishing gradient issues, while dropout layers reduce overfitting during training. T1, T2, and FLAIR modalities from the BraTS 2019 and 2020 datasets were used to evaluate the model. Experimental results show high segmentation accuracy on BraTS 2020 and the Dice Similarity Coefficients (DSC) achieved were 0.9136 (WT), 0.7143 (TC), and 0.7028 (ET). The paper demonstrates that patch-wise deep residual architectures, even with limited training data, can deliver reliable and robust brain tumor segmentation.

Background Accurate monitoring of feeding behavior in grazing ruminants, particularly the detection of prehension events, is a central challenge for Precision Livestock Farming (PLF). Traditional methods, such as accelerometers, show limitations in the reliable identification of individual events. Acoustic analysis based on deep learning is emerging as a non-invasive and promising alternative. Methods This study presents two main contributions: (i) a web-based software platform (built on React.js and TensorFlow.js) for the annotation, visualization, and in-browser inference of audio signals; (ii) a comparative analysis of several 2D-CNN architectures (DenseNet-121, ResNet-101, EfficientNet-B7, and YOLO11s-cls) for the classification of prehension events. Models were trained and tested on a dataset of logarithmic spectrograms (500 ms) derived from audio recordings acquired via collars on cattle. Results Analysis revealed high performance across all architectures. Although DenseNet-121 achieved the highest weighted metrics (Accuracy 83.7%, AUC 0.90), the YOLO11s-cls model demonstrated remarkable competitiveness, achieving nearly identical accuracy (83.1%) but with significantly superior computational efficiency (4.5 ms inference time). Crucially for field applications, YOLO exhibited excellent rejection of non-relevant sounds, with a 91% Specificity on the “no-prehension” class. Conclusions The study validates the efficacy of spectrogram-based 2D-CNNs for ingestion monitoring and identifies YOLO as a promising candidate for efficiency-oriented deployment scenarios, offering a favorable trade-off between predictive reliability and low-latency requirements. The developed platform further supports this transition from research to in-field application.

Introduction Over the past decade, there has been a significant increase in studies using biomedical signals for objective monitoring of Parkinson’s disease (PD) motor symptoms. Inertial sensors are widely employed to record motion, producing time-series data that capture the underlying motor condition of patients. A major challenge in the field is classifying these signals to discriminate healthy subjects from PD individuals and distinguish motor conditions among patients. While many studies focus on feature classification, there is a lack of research on the influence of signal preprocessing. Methods To fill this gap, we evaluate data from healthy subjects and PD patients during interaction with the RehaBEElitation serious game. We employed the catch22 feature set to extract robust time-series characteristics. To evaluate the influence of preprocessing on classification between healthy individuals and patients in on and off medication states, four strategies were adopted. Results Initially, features extracted from raw data showed limited accuracy due to noise and voluntary movements. Subsequent interpolation to address discontinuities produced inconsistent results. The third strategy involved wavelet decomposition, which effectively mitigated trends and motion artifacts, resulting in a significant increase in accuracy across all models and confirming the vital role of sophisticated signal filtering. The fourth strategy combined interpolation and wavelet decomposition, achieving the best results with optimal separation (Accuracy = 100.0%) in binary classification and significant improvement in the multi-class problem. Discussion Our findings establish that signal conditioning is pivotal for maximizing discriminative power. To further validate our findings, we benchmarked our pipeline against the RandOm Convolutional KErnel Transform (ROCKET) using a RidgeClassifierCV. The catch22 with Random Forest (RF) classifier, using a wavelet-based approach, achieved a balanced accuracy of 76.0% in the multiclass task, demonstrating superior performance compared to the ROCKETRidgeClassifierCV framework (69.0%) while maintaining a more compact and computationally efficient feature representation.

Face recognition is currently one of the most popular forms of biometric verification. As the effectiveness and security of this solution increase, so does its use in specialized fields. Considering that the verification process involves thousands of people, often under varying lighting conditions and with equipment of different parameters, biometric samples are of mixed quality. Therefore, there is a need to define the conditions under which a biometric sample is objectively good for a face recognition system. To address this, the international standard ISO/IEC 29794-5:2025 was developed, with defined quality measures, along with a description of suggested implementation where the majority of substantive work has already been completed. The aim of this work is to provide non-proprietary implementation of the ISO/IEC 29794-5:2023 standard for face image quality assessment and to compare its performance against OFIQ reference implementation. More broadly, this study examines the common challenge that biometric standards sometimes propose ideas that may not be top-effective in real-life operational scenarios. This paper includes the implementation of two systems for assessing face image quality based on selected standard’s measures. The first system follows the implementation suggested directly by the standard, while the second utilizes the latest scientific and commercial solutions. Ultimately, these systems are compared using a database of photographs differentiated by demographics and quality.

Introduction Driving a powered wheelchair is a complex task that requires the integration of motor, visual, and cognitive skills. The development of assistive technologies without appropriate assessment methods that help bridge the gap between users and developers may lead to abandonment and reduced engagement. Most assessments rely on explicit measures, such as performance metrics, or subjective tools like interviews and questionnaires. In contrast, implicit measures allow continuous inference of mental states during task execution. This study proposes the use of blink indices derived from electroencephalographic (EEG) signals as implicit metrics to estimate cognitive load during the use of a virtual reality wheelchair training simulator. Methods A total of 25 participants (14 females and 11 males; mean age 26.50 ± 5.7 years) completed a predefined route using a virtual wheelchair simulator. Blink parameters, including frequency, duration, and velocity, were extracted from EEG signals during task performance. After completing the simulation, participants responded to the NASA Task Load Index (NASA-TLX) to assess subjective cognitive load, as well as the System Usability Scale (SUS) and the Igroup Presence Questionnaire (IPQ). Results The findings showed that higher mental-visual demand was associated with decreases in blink frequency, duration, and velocity. Correlation analyses between NASA-TLX scores and blink parameters revealed weak to moderate associations. These results suggest partial convergence between subjective and physiological measures of cognitive load. Discussion Blink-based indices derived from EEG signals provide relevant information regarding cognitive demand during wheelchair simulator use. However, blink parameters alone are insufficient to reliably infer cognitive load. When combined with subjective questionnaires, implicit physiological metrics may offer a more comprehensive assessment than questionnaires alone, supporting the development and refinement of assistive training technologies.

The integration of unmanned aerial vehicles (UAVs) into precision agriculture, as envisioned in the agricultural systems, promises significant gains in crop monitoring, yield forecasting, and targeted agro-technical interventions. However, the use of IMT frequency bands for real-time UAV communications introduces new spectrum sharing and compatibility challenges. Unlike terrestrial user equipment, airborne agricultural drones operate always outdoors, above the base-station downtilt, with predominantly line-of-sight (LoS) propagation to multiple cells, drastically altering compatibility conditions and potentially increasing interference to other operators. This paper proposes a Monte Carlo-based simulation framework analysis of interference generated by such UAVs in IMT frequency allocations across 694–2690 MHz. Simulations model rural and urban macrocell deployments typical of large-scale farmlands, incorporating 3D antenna patterns, altitude-dependent air-to-ground channel models, realistic LTE/NR power-control schemes, and UAV operational patterns. Key metrics include aggregate uplink interference at victim cells, downlink degradation at UAVs, and cross-link interference in TDD systems. Results show that even low-power UAV transmissions can exceed harmful interference thresholds in multiple adjacent-channel cells. Operational recommendations are provided to ensure coexistence of precision-agriculture UAVs with terrestrial IMT networks.

Integrated Sensing and Communication (ISAC) is rapidly evolving from a forward-looking concept into a practical design principle for modern wireless systems. This Research Topic focuses on opportunistic ISAC, where sensing functionality is extracted from existing communication infrastructure, including commercial microwave links (CMLs), satellite channels, and automotive radars, to monitor the environment without deploying dedicated, costly hardware. By treating ambient radio signals as "opportunistic" sensors, one can retrieve high-resolution environmental information from networks that already exist. This capabilities expands the capabilities of the standard ISAC paradigm, as it uses an already deployed communication hardware, which was not originally designed for ISAC.Environmental information is often already embedded in radio measurements. For example, rainfall and water vapor affect microwave signlas, and can cause extra attenuation on microwave links channels. By processing the attenuation of these links, rainfall and water vapor monitoring can be realized. In addition, mmWave sensing also captures aspects of the physical environment, where robust processing of sparse radar point clouds can reveal structured features such as road boundaries. Turning these diverse sources of information into reliable weather and environmental products requires careful modeling, robust detection and estimation under non-stationary conditions, and principled fusion with complementary sensors and data sources.The five articles in this Topic reflect these capabilities from a few different angles. They show how opportunistic measurements can be turned into rainfall and humidity products, how algorithmic choices can be made with the end metric in mind, and how mmWave sensing methods can be advantageous for smart-city monitoring and transportation safety. In this short editorial, we summarize each article in turn and discuss its context and importance within the broader field.

The brain–computer interface (BCI) is the connection between the human brain and computers, creating a bridge that mimics the human brain. The premise behind near-infrared spectroscopy (NIRS) is that increased oxygen consumption in the brain leads to increased blood flow due to nerve connections. NIRS is a non-invasive procedure; changes in oxyhemoglobin (Oxy-Hb) and deoxyhemoglobin (Deoxy-Hb) parameters can be easily utilized to detect brain hemodynamics. This study is based on the Oxy-Hb parameter to classify mental arithmetic and rest states of the brain using singular spectrum analysis (SSA). SSA results in a better-denoised signal and decomposition into different principal components for analysis of these states. Oxy- and Deoxy-Hb patterns are temporary and unstable, so features such as power bandwidth, entropy, and complexity were extracted for classification. The reported accuracy in existing methods is 79.4% for the antagonistic single-trial classification and 86.9% for graph NIRS methods. The present study’s mean accuracy was 98.4% based on a set of selected features using filtering detrending (FD)-SSA, thus reducing the cost of poor sorting. Finally, classification models were evaluated based on scores such as Matthew’s correlation coefficient, precision, F1-score, and recall, resulting in 0.889, 0.968, 0.966, and 0.963, respectively.