Exploring the Possibility of Photoplethysmography-Based Human Activity Recognition Using Convolutional Neural Networks.

Various sensing modalities, including external and internal sensors, have been employed in research on human activity recognition (HAR). Among these, internal sensors, particularly wearable technologies, hold significant promise due to their lightweight nature and simplicity. Recently, HAR techniques leveraging wearable biometric signals, such as electrocardiography (ECG) and photoplethysmography (PPG), have been proposed using publicly available datasets. However, to facilitate broader practical applications, a more extensive analysis based on larger databases with cross-subject validation is required. In pursuit of this objective, we initially gathered PPG signals from 40 participants engaged in five common daily activities. Subsequently, we evaluated the feasibility of classifying these activities using deep learning architecture. The model's performance was assessed in terms of accuracy, precision, recall, and F-1 measure via cross-subject cross-validation (CV). The proposed method successfully distinguished the five activities considered, with an average test accuracy of 95.14%. Furthermore, we recommend an optimal window size based on a comprehensive evaluation of performance relative to the input signal length. These findings confirm the potential for practical HAR applications based on PPG and indicate its prospective extension to various domains, such as healthcare or fitness applications, by concurrently analyzing behavioral and health data through a single biometric signal.

Optimizing the maximum reported cluster size for the multinomial-based spatial scan statistic.

BACKGROUND: Correctly identifying spatial disease cluster is a fundamental concern in public health and epidemiology. The spatial scan statistic is widely used for detecting spatial disease clusters in spatial epidemiology and disease surveillance. Many studies default to a maximum reported cluster size (MRCS) set at 50% of the total population when searching for spatial clusters. However, this default setting can sometimes report clusters larger than true clusters, which include less relevant regions. For the Poisson, Bernoulli, ordinal, normal, and exponential models, a Gini coefficient has been developed to optimize the MRCS. Yet, no measure is available for the multinomial model. RESULTS: We propose two versions of a spatial cluster information criterion (SCIC) for selecting the optimal MRCS value for the multinomial-based spatial scan statistic. Our simulation study suggests that SCIC improves the accuracy of reporting true clusters. Analysis of the Korea Community Health Survey (KCHS) data further demonstrates that our method identifies more meaningful small clusters compared to the default setting. CONCLUSIONS: Our method focuses on improving the performance of the spatial scan statistic by optimizing the MRCS value when using the multinomial model. In public health and disease surveillance, the proposed method can be used to provide more accurate and meaningful spatial cluster detection for multinomial data, such as disease subtypes.

Evaluating the Window Size's Role in Automatic EEG Epilepsy Detection.

Electroencephalography is one of the most commonly used methods for extracting information about the brain's condition and can be used for diagnosing epilepsy. The EEG signal's wave shape contains vital information about the brain's state, which can be challenging to analyse and interpret by a human observer. Moreover, the characteristic waveforms of epilepsy (sharp waves, spikes) can occur randomly through time. Considering all the above reasons, automatic EEG signal extraction and analysis using computers can significantly impact the successful diagnosis of epilepsy. This research explores the impact of different window sizes on EEG signals' classification accuracy using four machine learning classifiers. The machine learning methods included a neural network with ten hidden nodes trained using three different training algorithms and the k-nearest neighbours classifier. The neural network training methods included the Broyden-Fletcher-Goldfarb-Shanno algorithm, the multistart method for global optimization problems, and a genetic algorithm. The current research utilized the University of Bonn dataset containing EEG data, divided into epochs having 50% overlap and window lengths ranging from 1 to 24 s. Then, statistical and spectral features were extracted and used to train the above four classifiers. The outcome from the above experiments showed that large window sizes with a length of about 21 s could positively impact the classification accuracy between the compared methods.

An Entropy Analysis-Based Window Size Optimization Scheme for Merging LiDAR Data Frames.

LiDAR is a useful technology for gathering point cloud data from its environment and has been adapted to many applications. We use a cost-efficient LiDAR system attached to a moving object to estimate the location of the moving object using referenced linear structures. In the stationary state, the accuracy of extracting linear structures is low given the low-cost LiDAR. We propose a merging scheme for the LiDAR data frames to improve the accuracy by using the movement of the moving object. The proposed scheme tries to find the optimal window size by means of an entropy analysis. The optimal window size is determined by finding the minimum point between the entropy indicator of the ideal result and the entropy indicator of the actual result of each window size. The proposed indicator can describe the accuracy of the entire path of the moving object at each window size using a simple single value. The experimental results show that the proposed scheme can improve the linear structure extraction accuracy.

Decoding lexical tones and vowels in imagined tonal monosyllables using fNIRS signals.

Objective.Speech is a common way of communication. Decoding verbal intent could provide a naturalistic communication way for people with severe motor disabilities. Active brain computer interaction (BCI) speller is one of the most commonly used speech BCIs. To reduce the spelling time of Chinese words, identifying vowels and tones that are embedded in imagined Chinese words is essential. Functional near-infrared spectroscopy (fNIRS) has been widely used in BCI because it is portable, non-invasive, safe, low cost, and has a relatively high spatial resolution.Approach.In this study, an active BCI speller based on fNIRS is presented by covertly rehearsing tonal monosyllables with vowels (i.e. /a/, /i/, /o/, and /u/) and four lexical tones in Mandarin Chinese (i.e. tones 1, 2, 3, and 4) for 10 s.Main results.fNIRS results showed significant differences in the right superior temporal gyrus between imagined vowels with tone 2/3/4 and those with tone 1 (i.e. more activations and stronger connections to other brain regions for imagined vowels with tones 2/3/4 than for those with tone 1). Speech-related areas for tone imagery (i.e. the right hemisphere) provided majority of information for identifying tones, while the left hemisphere had advantages in vowel identification. Having decoded both vowels and tones during the post-stimulus 15 s period, the average classification accuracies exceeded 40% and 70% in multiclass (i.e. four classes) and binary settings, respectively. To spell words more quickly, the time window size for decoding was reduced from 15 s to 2.5 s while the classification accuracies were not significantly reduced.Significance.For the first time, this work demonstrated the possibility of discriminating lexical tones and vowels in imagined tonal syllables simultaneously. In addition, the reduced time window for decoding indicated that the spelling time of Chinese words could be significantly reduced in the fNIRS-based BCIs.

Improved pig behavior analysis by optimizing window sizes for individual behaviors on acceleration and angular velocity data.

This paper presents the application of machine learning algorithms to identify pigs' behaviors from data collected using the wireless sensor nodes mounted on pigs. The sensor node attached to a pig's back senses the acceleration and angular velocity in three axes, and the sensed data are transmitted to a host computer wirelessly. Two video cameras, one attached to the ceiling of the pigpen and the other one to a fence, provided ground truth for data annotations. The data were collected from pigs for 131 h over 2 mo. As the typical behavior period depends on the behavior type, we segmented the acceleration data with different window sizes (WS) and step sizes (SS), and tested how the classification performance of different activities varied with different WS and SS. After exploring the possible combinations, we selected the optimum WS and SS. To compare performance, we used five machine learning algorithms, specifically support vector machine, k-nearest neighbors, decision trees, naive Bayes, and random forest (RF). Among the five algorithms, RF achieved the highest F1 score for four major behaviors consisting of 92.36% in total. The F1 scores of the algorithm were 0.98 for "eating," 0.99 for "lying," 0.93 for "walking," and 0.91 for "standing" behaviors. The optimal WS was 7 s for "eating" and "lying," and 3 s for "walking" and "standing." The proposed work demonstrates that, based on the length of behavior, the adaptive window and step sizes increase the classification performance.

Analyzing the behavior of pigs provides great insights into animal welfare and health. Technologies that enable automatic, continuous, and real-time behavior monitoring have emerged as alternative solutions and have received considerable attention. Using sensor-based animal monitoring technology, we could provide objective and quantitative assessments of the health and continuous care of pigs. We extracted distinct characteristics/features of different activities over given segments of acceleration data to boost classification performance. Pigs have various behavior patterns with different durations; treating behaviors with a small duration the same as a long duration could ignore the minority behaviors in the window frame. Our study showed that by finding the adaptive window sizes customized for individual behaviors, we could reduce the chance of mixing activities and compute the feature for better classification performance.

Data augmentation and deep neural networks for the classification of Pakistani racial speakers recognition.

Speech emotion recognition (SER) systems have evolved into an important method for recognizing a person in several applications, including e-commerce, everyday interactions, law enforcement, and forensics. The SER system's efficiency depends on the length of the audio samples used for testing and training. However, the different suggested models successfully obtained relatively high accuracy in this study. Moreover, the degree of SER efficiency is not yet optimum due to the limited database, resulting in overfitting and skewing samples. Therefore, the proposed approach presents a data augmentation method that shifts the pitch, uses multiple window sizes, stretches the time, and adds white noise to the original audio. In addition, a deep model is further evaluated to generate a new paradigm for SER. The data augmentation approach increased the limited amount of data from the Pakistani racial speaker speech dataset in the proposed system. The seven-layer framework was employed to provide the most optimal performance in terms of accuracy compared to other multilayer approaches. The seven-layer method is used in existing works to achieve a very high level of accuracy. The suggested system achieved 97.32% accuracy with a 0.032% loss in the 75%:25% splitting ratio. In addition, more than 500 augmentation data samples were added. Therefore, the proposed approach results show that deep neural networks with data augmentation can enhance the SER performance on the Pakistani racial speech dataset.

Assessing Impact of Sensors and Feature Selection in Smart-Insole-Based Human Activity Recognition.

Human Activity Recognition (HAR) is increasingly used in a variety of applications, including health care, fitness tracking, and rehabilitation. To reduce the impact on the user's daily activities, wearable technologies have been advanced throughout the years. In this study, an improved smart insole-based HAR system is proposed. The impact of data segmentation, sensors used, and feature selection on HAR was fully investigated. The Support Vector Machine (SVM), a supervised learning algorithm, has been used to recognise six ambulation activities: downstairs, sit to stand, sitting, standing, upstairs, and walking. Considering the impact that data segmentation can have on the classification, the sliding window size was optimised, identifying the length of 10 s with 50% of overlap as the best performing. The inertial sensors and pressure sensors embedded into the smart insoles have been assessed to determine the importance that each one has in the classification. A feature selection technique has been applied to reduce the number of features from 272 to 227 to improve the robustness of the proposed system and to investigate the importance of features in the dataset. According to the findings, the inertial sensors are reliable for the recognition of dynamic activities, while pressure sensors are reliable for stationary activities; however, the highest accuracy (94.66%) was achieved by combining both types of sensors.

Continuous Classification of Locomotion in Response to Task Complexity and Anticipatory State.

OBJECTIVE: Intent recognition in lower-extremity assistive devices (e.g., prostheses and exoskeletons) is typically limited to either recognition of steady-state locomotion or changes of terrain (e.g., level ground to stair) occurring in a straight-line path and under anticipated condition. Stability is highly affected during non-steady changes of direction such as cuts especially when they are unanticipated, posing high risk of fall-related injuries. Here, we studied the influence of changes of direction and user anticipation on task recognition, and accordingly introduced classification schemes accommodating such effects. METHODS: A linear discriminant analysis (LDA) classifier continuously classified straight-line walking, sidestep/crossover cuts (single transitions), and cuts-to-stair locomotion (mixed transitions) performed under varied task anticipatory conditions. Training paradigms with varying levels of anticipated/unanticipated exposures and analysis windows of size 100-600 ms were examined. RESULTS: More accurate classification of anticipated relative to unanticipated tasks was observed. Including bouts of target task in the training data was necessary to improve generalization to unanticipated locomotion. Only up to two bouts of target task were sufficient to reduce errors to <20% in unanticipated mixed transitions, whereas, in single transitions and straight walking, substantial unanticipated information (i.e., five bouts) was necessary to achieve similar outcomes. Window size modifications did not have a significant influence on classification performance. CONCLUSION: Adjusting the training paradigm helps to achieve classification schemes capable of adapting to changes of direction and task anticipatory state. SIGNIFICANCE: The findings could provide insight into developing classification schemes that can adapt to changes of direction and user anticipation. They could inform intent recognition strategies for controlling lower-limb assistive to robustly handle "unknown" circumstances, and thus deliver increased level of reliability and safety.

A Comparative Study of Window Size and Channel Arrangement on EEG-Emotion Recognition Using Deep CNN.

Emotion recognition based on electroencephalograms has become an active research area. Yet, identifying emotions using only brainwaves is still very challenging, especially the subject-independent task. Numerous studies have tried to propose methods to recognize emotions, including machine learning techniques like convolutional neural network (CNN). Since CNN has shown its potential in generalization to unseen subjects, manipulating CNN hyperparameters like the window size and electrode order might be beneficial. To our knowledge, this is the first work that extensively observed the parameter selection effect on the CNN. The temporal information in distinct window sizes was found to significantly affect the recognition performance, and CNN was found to be more responsive to changing window sizes than the support vector machine. Classifying the arousal achieved the best performance with a window size of ten seconds, obtaining 56.85% accuracy and a Matthews correlation coefficient (MCC) of 0.1369. Valence recognition had the best performance with a window length of eight seconds at 73.34% accuracy and an MCC value of 0.4669. Spatial information from varying the electrode orders had a small effect on the classification. Overall, valence results had a much more superior performance than arousal results, which were, perhaps, influenced by features related to brain activity asymmetry between the left and right hemispheres.

A new focus measure operator for enhancing image focus in 3D shape recovery.

Measuring the image focus is an important issue in Shape from Focus methods. Conventionally, the Sum of Modified Laplacian, Gray Level Variance (GLV), and Tenengrad techniques have been used frequently among various focus measure operators for estimating the focus levels in a sequence of images. However, they have various issues such as fixed window size and suboptimal focus quality. To solve these problems, a new focus measure operator based on the adaptive sum of weighted modified Laplacian is proposed. First, the adaptive window size selection algorithm based on the GLV is applied. Next, appropriate weights are assigned to the Modified Laplacian values in the image window based on the distance between the center pixel and neighboring pixels. Finally, the Weighted Modified Laplacian values in the image window are summed. Experimental results demonstrate the effectiveness of the proposed method.

Industry 4.0 towards Forestry 4.0: Fire Detection Use Case.

Forestry 4.0 is inspired by the Industry 4.0 concept, which plays a vital role in the next industrial generation revolution. It is ushering in a new era for efficient and sustainable forest management. Environmental sustainability and climate change are related challenges to promote sustainable forest management of natural resources. Internet of Forest Things (IoFT) is an emerging technology that helps manage forest sustainability and protect forest from hazards via distributing smart devices for gathering data stream during monitoring and detecting fire. Stream processing is a well-known research area, and recently, it has gained a further significance due to the emergence of IoFT devices. Distributed stream processing platforms have emerged, e.g., Apache Flink, Storm, and Spark, etc. Querying windowing is the heart of any stream-processing platform which splits infinite data stream into chunks of finite data to execute a query. Dynamic query window-based processing can reduce the reporting time in case of missing and delayed events caused by data drift.In this paper, we present a novel dynamic mechanism to recommend the optimal window size and type based on the dynamic context of IoFT application. In particular, we designed a dynamic window selector for stream queries considering input stream data characteristics, application workload and resource constraints to recommend the optimal stream query window configuration. A research gap on the likelihood of adopting smart IoFT devices in environmental sustainability indicates a lack of empirical studies to pursue forest sustainability, i.e., sustainable forestry applications. So, we focus on forest fire management and detection as a use case of Forestry 4.0, one of the dynamic environmental management challenges, i.e., climate change, to deliver sustainable forestry goals. According to the dynamic window selector's experimental results, end-to-end latency time for the reported fire alerts has been reduced by dynamical adaptation of window size with IoFT stream rate changes.

A high-performance approach for predicting donor splice sites based on short window size and imbalanced large samples.

BACKGROUND: Splice sites prediction has been a long-standing problem in bioinformatics. Although many computational approaches developed for splice site prediction have achieved satisfactory accuracy, further improvement in predictive accuracy is significant, for it is contributing to predict gene structure more accurately. Determining a proper window size before prediction is necessary. Overly long window size may introduce some irrelevant features, which would reduce predictive accuracy, while the use of short window size with maximum information may performs better in terms of predictive accuracy and time cost. Furthermore, the number of false splice sites following the GT-AG rule far exceeds that of true splice sites, accurate and rapid prediction of splice sites using imbalanced large samples has always been a challenge. Therefore, based on the short window size and imbalanced large samples, we developed a new computational method named chi-square decision table (χ2-DT) for donor splice site prediction. RESULTS: Using a short window size of 11 bp, χ2-DT extracts the improved positional features and compositional features based on chi-square test, then introduces features one by one based on information gain, and constructs a balanced decision table aimed at implementing imbalanced pattern classification. With a 2000:271,132 (true sites:false sites) training set, χ2-DT achieves the highest independent test accuracy (93.34%) when compared with three classifiers (random forest, artificial neural network, and relaxed variable kernel density estimator) and takes a short computation time (89 s). χ2-DT also exhibits good independent test accuracy (92.40%), when validated with BG-570 mutated sequences with frameshift errors (nucleotide insertions and deletions). Moreover, χ2-DT is compared with the long-window size-based methods and the short-window size-based methods, and is found to perform better than all of them in terms of predictive accuracy. CONCLUSIONS: Based on short window size and imbalanced large samples, the proposed method not only achieves higher predictive accuracy than some existing methods, but also has high computational speed and good robustness against nucleotide insertions and deletions. REVIEWERS: This article was reviewed by Ryan McGinty, Ph.D. and Dirk Walther.

Assessment of dynamic functional connectivity in resting-state fMRI using the sliding window technique.

INTRODUCTION: Recent studies related to assessing functional connectivity (FC) in resting-state functional magnetic resonance imaging have revealed that the resulting connectivity patterns exhibit considerable fluctuations (dynamic FC [dFC]). A widely applied method for quantifying dFC is the sliding window technique. According to this method, the data are divided into segments with the same length (window size) and a correlation metric is employed to assess the connectivity within these segments, whereby the window size is often empirically chosen. METHODS: In this study, we rigorously investigate the assessment of dFC using the sliding window approach. Specifically, we perform a detailed comparison between different correlation metrics, including Pearson, Spearman and Kendall correlation, Pearson and Spearman partial correlation, Mutual Information (MI), Variation of Information (VI), Kullback-Leibler divergence, Multiplication of Temporal Derivatives and Inverse Covariance. RESULTS: Using test-retest datasets, we show that MI and VI yielded the most consistent results by achieving high reliability with respect to dFC estimates for different window sizes. Subsequent hypothesis testing, based on multivariate phase randomization surrogate data generation, allowed the identification of dynamic connections between the posterior cingulate cortex and regions in the frontal lobe and inferior parietal lobes, which were overall in agreement with previous studies. CONCLUSIONS: In the case of MI and VI, a window size of at least 120 s was found to be necessary for detecting dFC for some of the previously identified dynamically connected regions.

Vehicle Mode and Driving Activity Detection Based on Analyzing Sensor Data of Smartphones.

In this paper, we present a flexible combined system, namely the Vehicle mode-driving Activity Detection System (VADS), that is capable of detecting either the current vehicle mode or the current driving activity of travelers. Our proposed system is designed to be lightweight in computation and very fast in response to the changes of travelers' vehicle modes or driving events. The vehicle mode detection module is responsible for recognizing both motorized vehicles, such as cars, buses, and motorbikes, and non-motorized ones, for instance, walking, and bikes. It relies only on accelerometer data in order to minimize the energy consumption of smartphones. By contrast, the driving activity detection module uses the data collected from the accelerometer, gyroscope, and magnetometer of a smartphone to detect various driving activities, i.e., stopping, going straight, turning left, and turning right. Furthermore, we propose a method to compute the optimized data window size and the optimized overlapping ratio for each vehicle mode and each driving event from the training datasets. The experimental results show that this strategy significantly increases the overall prediction accuracy. Additionally, numerous experiments are carried out to compare the impact of different feature sets (time domain features, frequency domain features, Hjorth features) as well as the impact of various classification algorithms (Random Forest, Naïve Bayes, Decision tree J48, K Nearest Neighbor, Support Vector Machine) contributing to the prediction accuracy. Our system achieves an average accuracy of 98.33% in detecting the vehicle modes and an average accuracy of 98.95% in recognizing the driving events of motorcyclists when using the Random Forest classifier and a feature set containing time domain features, frequency domain features, and Hjorth features. Moreover, on a public dataset of HTC company in New Taipei, Taiwan, our framework obtains the overall accuracy of 97.33% that is considerably higher than that of the state-of the art.

Evaluation of Bias Correction Method for Satellite-Based Rainfall Data.

With the advances in remote sensing technology, satellite-based rainfall estimates are gaining attraction in the field of hydrology, particularly in rainfall-runoff modeling. Since estimates are affected by errors correction is required. In this study, we tested the high resolution National Oceanic and Atmospheric Administration's (NOAA) Climate Prediction Centre (CPC) morphing technique (CMORPH) satellite rainfall product (CMORPH) in the Gilgel Abbey catchment, Ethiopia. CMORPH data at 8 km-30 min resolution is aggregated to daily to match in-situ observations for the period 2003-2010. Study objectives are to assess bias of the satellite estimates, to identify optimum window size for application of bias correction and to test effectiveness of bias correction. Bias correction factors are calculated for moving window (MW) sizes and for sequential windows (SW's) of 3, 5, 7, 9, …, 31 days with the aim to assess error distribution between the in-situ observations and CMORPH estimates. We tested forward, central and backward window (FW, CW and BW) schemes to assess the effect of time integration on accumulated rainfall. Accuracy of cumulative rainfall depth is assessed by Root Mean Squared Error (RMSE). To systematically correct all CMORPH estimates, station based bias factors are spatially interpolated to yield a bias factor map. Reliability of interpolation is assessed by cross validation. The uncorrected CMORPH rainfall images are multiplied by the interpolated bias map to result in bias corrected CMORPH estimates. Findings are evaluated by RMSE, correlation coefficient (r) and standard deviation (SD). Results showed existence of bias in the CMORPH rainfall. It is found that the 7 days SW approach performs best for bias correction of CMORPH rainfall. The outcome of this study showed the efficiency of our bias correction approach.

Varying behavior of different window sizes on the classification of static and dynamic physical activities from a single accelerometer.

Accuracy of systems able to recognize in real time daily living activities heavily depends on the processing step for signal segmentation. So far, windowing approaches are used to segment data and the window size is usually chosen based on previous studies. However, literature is vague on the investigation of its effect on the obtained activity recognition accuracy, if both short and long duration activities are considered. In this work, we present the impact of window size on the recognition of daily living activities, where transitions between different activities are also taken into account. The study was conducted on nine participants who wore a tri-axial accelerometer on their waist and performed some short (sitting, standing, and transitions between activities) and long (walking, stair descending and stair ascending) duration activities. Five different classifiers were tested, and among the different window sizes, it was found that 1.5 s window size represents the best trade-off in recognition among activities, with an obtained accuracy well above 90%. Differences in recognition accuracy for each activity highlight the utility of developing adaptive segmentation criteria, based on the duration of the activities.

Synthesis of large-pore mesostructured cellular foam silica spheres for the adsorption of biomolecules.

Mesostructured cellular foam (MCF) silica spheres with different textual parameters were synthesized using a triblock copolymer as a template. The effects of acid concentration and aging time on the window size and morphology were discussed. Besides, the adsorption performances of lysozyme and bovine serum albumin on the blank MCF with different window size and aminopropyl-modified MCFs were studied. The adsorption capacity and rates were shown to be dependent on the window size and surface chemical properties of the adsorbents. In particular, the MCF with window size of 15.2 nm showed fast adsorption for lysozyme with an adsorption capacity of 500 mg/g in 10 min. Furthermore, it has been shown that MCF spheres are potential materials in the separation of biomolecules because of their chemical tunable surface and molecular sieve properties.

A pulse artifact removal method considering artifact variations in the simultaneous recording of EEG and fMRI.

A simultaneous recording of electroencephalogram (EEG) and functional magnetic resonance imaging (fMRI) can provide high spatiotemporal information of brain activity. However, a proper analysis of the EEG signals is often hindered by various artifacts. In particular, pulse artifact (PA) induced from the heartbeat of a subject interferes with reliable measurements of the EEG signal. A new PA removal method that takes into account the delay variation between the heartbeat and PA and the window size variation in PA is presented in order to improve the detection and suppression of PA in EEG signals. A PA is classified into either a normal PA or a deformed PA. Only normal PAs are averaged to generate a PA template that is used to remove PAs from the measured EEG signals. The performance of the proposed method was evaluated by simulated data and real EEG measurements from epilepsy patients. The results are compared with those from conventional methods.