Using EEG-EEG or EEG-ECG transfer learning, this study explored the potential of training fundamental cross-domain convolutional neural networks (CNNs) for applications in seizure prediction and sleep staging, respectively. Notwithstanding the seizure model's identification of interictal and preictal periods, the sleep staging model classified signals into five distinct stages. Successfully personalizing a seizure prediction model with six frozen layers, the model achieved 100% accuracy for seven out of nine patients in just 40 seconds of training time. Importantly, the cross-signal transfer learning EEG-ECG model for sleep staging displayed an accuracy approximately 25% greater than the ECG-alone model; concurrently, training time was reduced by more than half. Personalized EEG signal models, generated through transfer learning from existing models, contribute to both quicker training and heightened accuracy, consequently overcoming hurdles related to data inadequacy, variability, and inefficiencies.
Indoor locations, lacking sufficient air exchange, are prone to contamination by hazardous volatile compounds. For the purpose of minimizing associated risks, monitoring the distribution of indoor chemicals is highly important. A machine learning-driven monitoring system is introduced to process the data from a low-cost, wearable volatile organic compound (VOC) sensor used in a wireless sensor network (WSN). Localization of mobile devices in the WSN network is achieved through the use of fixed anchor nodes. The chief difficulty in deploying mobile sensor units for indoor applications is achieving their precise localization. Most definitely. check details Mobile device localization was performed by implementing machine learning algorithms on received signal strength indicators (RSSIs), pinpointing their source on a predefined map. A 120 square meter indoor location with a meandering path exhibited localization accuracy greater than 99%, as shown by the tests conducted. Utilizing a commercially available metal oxide semiconductor gas sensor, the WSN was deployed to map the distribution of ethanol originating from a point source. A PhotoIonization Detector (PID) measurement of ethanol concentration showed a correlation with the sensor signal, thereby demonstrating the simultaneous localization and detection of the volatile organic compound (VOC) source.
The burgeoning field of sensor and information technology has facilitated machines' ability to recognize and decipher human emotional states. Emotion recognition research holds considerable importance within various academic and practical domains. Human emotions display themselves in a wide range of forms. Therefore, the determination of emotions is attainable through analysis of facial expressions, spoken words, actions, or physiological metrics. Different sensors are used to collect these signals. Precisely discerning human emotional states fosters the growth of affective computing technologies. Existing emotion recognition surveys primarily rely on data from a single sensor. Hence, a crucial aspect is the comparison of diverse sensors, encompassing both unimodal and multimodal approaches. Employing a thorough review of the literature, this survey scrutinizes in excess of 200 papers on the topic of emotion recognition. These papers are categorized by the variations in the innovations they introduce. These articles' focus is on the employed methods and datasets for emotion recognition utilizing diverse sensor platforms. The survey also explores diverse uses and the most recent progress in the area of emotion recognition. This survey, furthermore, evaluates the strengths and limitations of diverse sensor technologies in emotion recognition. The proposed survey is designed to enhance researchers' comprehension of existing emotion recognition systems, ultimately improving the selection of appropriate sensors, algorithms, and datasets.
Our proposed approach to designing ultra-wideband (UWB) radar utilizes pseudo-random noise (PRN) sequences. Its crucial characteristics encompass user-tailorable capabilities for diverse microwave imaging applications, and its potential for multichannel scaling. A fully synchronized multichannel radar imaging system, designed for short-range imaging tasks like mine detection, non-destructive testing (NDT), or medical imaging, is presented through its advanced system architecture. Emphasis is placed on the implemented synchronization mechanism and clocking scheme. The targeted adaptivity's core functionality is implemented through hardware, encompassing variable clock generators, dividers, and programmable PRN generators. An extensive open-source framework, present within the Red Pitaya data acquisition platform, enables the customization of signal processing, in addition to enabling the utilization of adaptive hardware. The attainable performance of the implemented prototype system is measured by a system benchmark that scrutinizes signal-to-noise ratio (SNR), jitter, and the stability of synchronization. Moreover, an assessment of the envisioned future progress and enhancement of performance is detailed.
Precise point positioning in real-time relies heavily on the performance of ultra-fast satellite clock bias (SCB) products. This paper aims to enhance the predictive capability of SCB within the Beidou satellite navigation system (BDS) by introducing a sparrow search algorithm to optimize the extreme learning machine (SSA-ELM), addressing the inadequacy of ultra-fast SCB for precise point positioning. Employing the sparrow search algorithm's robust global search and swift convergence, we enhance the predictive accuracy of the extreme learning machine's SCB. Employing ultra-fast SCB data from the international GNSS monitoring assessment system (iGMAS), this study carries out experiments. To gauge the precision and dependability of the data, the second-difference method is applied, confirming that the ultra-fast clock (ISU) products display an ideal match between observed (ISUO) and predicted (ISUP) data. The rubidium (Rb-II) and hydrogen (PHM) clocks on BDS-3 show superior accuracy and stability to those on BDS-2; this difference in reference clocks influences the accuracy of the SCB. SCB prediction was performed using SSA-ELM, quadratic polynomial (QP), and a grey model (GM), and the findings were compared to ISUP data. In predicting 3- and 6-hour outcomes utilizing 12 hours of SCB data, the SSA-ELM model demonstrably improves prediction accuracy, increasing prediction accuracy by approximately 6042%, 546%, and 5759% compared to the ISUP, QP, and GM models for 3-hour predictions, and 7227%, 4465%, and 6296% for 6-hour predictions, respectively. The accuracy of 6-hour predictions using 12 hours of SCB data is markedly improved by the SSA-ELM model, approximately 5316% and 5209% compared to the QP model, and 4066% and 4638% compared to the GM model. In closing, multiple-day data are instrumental in generating the 6-hour Short-Term Climate Bulletin (SCB) forecast. In light of the results, the predictive performance of the SSA-ELM model is enhanced by over 25% compared to the ISUP, QP, and GM models. The BDS-3 satellite achieves a greater degree of prediction accuracy than the BDS-2 satellite.
The crucial importance of human action recognition has driven considerable attention in the field of computer vision. Skeleton-sequence-based action recognition has seen significant advancement over the past decade. Conventional deep learning methods utilize convolutional operations to derive skeleton sequences. By learning spatial and temporal features through multiple streams, most of these architectures are realized. check details These investigations have broadened the understanding of action recognition through a multitude of algorithmic lenses. Nonetheless, three recurring challenges appear: (1) Models are commonly intricate, consequently necessitating a higher computational overhead. For supervised learning models, the dependence on labeled data during training is a persistent hindrance. The implementation of large models offers no real-time application benefit. Utilizing a multi-layer perceptron (MLP) with a contrastive learning loss function, dubbed ConMLP, this paper proposes a self-supervised learning framework to address the issues outlined above. ConMLP avoids the need for extensive computational resources, achieving impressive reductions in consumption. Supervised learning frameworks are often less adaptable to the massive datasets of unlabeled training data compared to ConMLP. Moreover, the system's requirements for configuration are low, allowing it to be readily incorporated into real-world applications. Results from extensive experiments on the NTU RGB+D dataset unequivocally place ConMLP at the top of the inference leaderboard, with a score of 969%. In comparison to the state-of-the-art self-supervised learning method, this accuracy is greater. Concomitantly, ConMLP is evaluated using a supervised learning paradigm, demonstrating recognition accuracy that matches or surpasses the leading methods.
Automated soil moisture management systems are common components of precision agricultural techniques. check details Although inexpensive sensors can significantly expand the spatial domain, this enhancement might be accompanied by a reduction in the accuracy of the data collected. This paper investigates the trade-offs between cost and accuracy in soil moisture sensing, contrasting low-cost and commercial sensors. Lab and field tests were conducted on the SKUSEN0193 capacitive sensor, forming the basis for the analysis. In conjunction with individual sensor calibration, two streamlined calibration methods are introduced: universal calibration utilizing all 63 sensors, and a single-point calibration leveraging soil sensor response in dry conditions. Sensor installation in the field, part of the second phase of testing, was carried out in conjunction with a low-cost monitoring station. Precipitation and solar radiation were the factors impacting the daily and seasonal oscillations in soil moisture, measurable by the sensors. Five factors—cost, accuracy, labor requirements, sample size, and life expectancy—were used to assess the performance of low-cost sensors in comparison to their commercial counterparts.