Tri-axial Accelerometer Data Research Articles

Deep learning based classification of sheep behaviour from accelerometer data with imbalance

Classification of sheep behaviour from a sequence of tri-axial accelerometer data has the potential to enhance sheep management. Sheep behaviour is inherently imbalanced (e.g., more ruminating than walking) resulting in underperforming classification for the minority activities which hold importance. Existing works have not addressed class imbalance and use traditional machine learning techniques, e.g., Random Forest (RF). We investigated Deep Learning (DL) models, namely, Long Short Term Memory (LSTM) and Bidirectional LSTM (BLSTM), appropriate for sequential data, from imbalanced data. Two data sets were collected in normal grazing conditions using jaw-mounted and ear-mounted sensors. Novel to this study, alongside typical single classes, e.g., walking, depending on the behaviours, data samples were labelled with compound classes, e.g., walking_grazing. The number of steps a sheep performed in the observed 10 s time window was also recorded and incorporated in the models. We designed several multi-class classification studies with imbalance being addressed using synthetic data. DL models achieved superior performance to traditional ML models, especially with augmented data (e.g., 4-Class + Steps: LSTM 88.0%, RF 82.5%). DL methods showed superior generalisability on unseen sheep (i.e., F1-score: BLSTM 0.84, LSTM 0.83, RF 0.65). LSTM, BLSTM and RF achieved sub-millisecond average inference time, making them suitable for real-time applications. The results demonstrate the effectiveness of DL models for sheep behaviour classification in grazing conditions. The results also demonstrate the DL techniques can generalise across different sheep. The study presents a strong foundation of the development of such models for real-time animal monitoring.

Can metrics of acceleration provide accurate estimates of energy costs of locomotion on uneven terrain? Using domestic sheep (Ovis aries) as an example

BackgroundLocomotion is often a necessity for animal survival and can account for a large proportion of an individual’s energy budget. Therefore, determining the energy costs of locomotion is an important part of understanding the interaction between an animal and its environment. Measures of animal acceleration, specifically ‘dynamic body acceleration’ (DBA) has proved to be a useful proxy of the energy cost of locomotion. However, few studies have considered the effects of interacting factors such as the animal’s speed or changes to the terrain slope on the putative acceleration versus energy expenditure relationship and how this may affect the relationship between DBA and energy expenditure.MethodsHere we conducted a methodological study to evaluate the ability of the metric ‘vectorial dynamic body acceleration’, VeDBA, obtained from tri-axial accelerometer data loggers, to act as a proxy for energy expenditure in non-uniform environments. We used indirect calorimetry to measure the oxygen consumption (V̇O2) of domestic sheep (Ovis aries) that were exposed to different ambient temperatures when immobile (resting) and that walked at various speeds (0.8 to 2.9 km h−1) and slope angles (− 6° to 6°) on a treadmill while simultaneously measuring tri-axial acceleration recorded at 40 Hz by body-mounted tags.ResultsThe lower critical temperature of sheep was identified as 18 °C, and V̇O2 when they were immobile was 3.67 mL O2 kg−1 min−1. There were positive relationships between V̇O2, VeDBA, and speed of walking. However, VeDBA correlated less well with V̇O2 when the terrain slope either inclined or declined.ConclusionsWe advocate caution when using DBA metrics for establishing energy use in animals moving over uneven terrain and suggest that each study species or location must be examined on a case-by-case basis. Reliance upon the relationship described between acceleration and energy expenditure on horizontal-surface treadmills can lead to potential under- or over-estimates of energy expenditure when animals walk on uneven or inclined ground.

A low-power HAR method for fall and high-intensity ADLs identification using wrist-worn accelerometer devices

Abstract There are many real-world applications like healthcare systems, job monitoring, well-being and personal fitness tracking, monitoring of elderly and frail people, assessment of rehabilitation and follow-up treatments, affording Fall Detection (FD) and ADL (Activity of Daily Living) identification, separately or even at a time. However, the two main drawbacks of these solutions are that most of the times, the devices deployed are obtrusive (devices worn on not quite common parts of the body like neck, waist and ankle) and the poor battery life. Thus, this work proposes a low-power classification algorithm based on an Ensemble of KNN and K-Means algorithms (EKMeans) to identify Falls and High-Intensity ADL events such as running, jogging and climbing up stairs. The input of EKMeans are triaxial accelerometer data gathered from wrist-wearable devices. The proposal will be validated on the Fall&ADL publicly available datasets UMAFall, UCIFall and FallAllD, considering two kinds of activity labelling: Two-Class and Multi-Class. An exhaustive comparative study between our proposal, and the baseline algorithms KNN and a feed-forward Neural Network (NN) is deployed, where EKMeans outperformed clearly the Specificity (ADL classification) of the KNN and NN for the three datasets. Finally, a comparative battery consumption study has been included deploying the analyzed algorithms in a WearOS smartwatch, where EKMeans drains the battery from 100% to 0% in 27.45 hours, saving 5% and 21% concerning KNN and NN, respectively. Keywords: Human Activity Recognition, ADL Identification, Fall Detection TS Clustering, TS Classification, Wearable Devices, Low-Power HAR.

A Cross-Sectional Study on the Characteristics of Physical Activity in Pre-Frail Older Adults.

This cross-sectional study aimed to characterize the physical activity (PA) of older adults with pre-frail status by examining sedentary behavior (SB) and PA using triaxial accelerometer data, with non-frail older adults as the control group. In this study, we divided the study participants into older adults who regularly participated in self-initiated citizen group exercise activities and those who did not. Data were collected between September and December 2017. We analyzed data from 256 older adults (women: 86.3%) aged ≥65 years. The interaction effect of participation status (participation and non-participation group) and frailty status (pre-frail and non-frail group) for moderate-to-vigorous PA (F = 9.178, p = 0.003) and daily mean number of steps (F = 9.351, p = 0.002) was significant. For the participation group, there was no difference between pre-frail older adults and non-frail older adults regarding length of SB and PA time, indicating that PA level was maintained in the participating pre-frail older adults. In contrast, moderate-to-vigorous PA and daily mean number of steps were low in pre-frail older adults who did not participate in the activities. The opportunity to participate in self-initiated group exercise activities and other PAs in the community may help pre-frail older adults maintain their PA.

Heat Stress Increases Movement Jerk During Physical Exertion.

Objective: Movement efficiency can be quantified during physical tasks by measuring the rate of change of acceleration (jerk). Jerk captures the smoothness of a motion and has been used to quantify movement for upper extremity and torso-based tasks. We collected triaxial accelerometer data during four physical tasks commonly performed in the work place to determine if jerk increases with physiologic strain.Methods: Participants completed a circuit of activities that mimicked the demands of manual labor in hot (40°C) and temperate (18°C) conditions. The circuit included walking on a treadmill carrying a load on the shoulder, lifting objects from the floor to the table, using a dead blow to strike the end of a heavy steel beam, and a kneeling rope pull. After the 9 min circuit, the participant had a standing rest for 1 min before repeating the circuit 3 additional times. Participants were instrumented with four 3-axis accelerometers (Actigraph wGT3X) secured to the torso, wrist, and upper arm.Results: There were 20 trials in the hot condition and 12 trials in the temperate condition. Heart rate and core body temperature increased during both protocols (p < 0.001). Measures of jerk varied by accelerometer location and activity. During treadmill walking, the wrist, torso, arm accelerometers measured higher jerk during the fourth circuit in the hot condition. During the lifting task, mean jerk increased in the hot condition in all accelerometers. Max jerk increased in the temperate condition in the arm accelerometer and jerk cost increased in the hot condition in the torso and arm accelerometers.Conclusions: Forty minutes of paced work performed in the heat resulted in increased acceleration and jerk in accelerometers placed on the torso, arm, and wrist. The accelerometers most consistently reporting these changes were task specific and suggest that a limited number of worn sensors could identify the onset of fatigue and increased injury risk.

What Are Sheep Doing? Tri-Axial Accelerometer Sensor Data Identify the Diel Activity Pattern of Ewe Lambs on Pasture.

Monitoring activity patterns of animals offers the opportunity to assess individual health and welfare in support of precision livestock farming. The purpose of this study was to use a triaxial accelerometer sensor to determine the diel activity of sheep on pasture. Six Perendale ewe lambs, each fitted with a neck collar mounting a triaxial accelerometer, were filmed during targeted periods of sheep activities: grazing, lying, walking, and standing. The corresponding acceleration data were fitted using a Random Forest algorithm to classify activity (=classifier). This classifier was then applied to accelerometer data from an additional 10 ewe lambs to determine their activity budgets. Each of these was fitted with a neck collar mounting an accelerometer as well as two additional accelerometers placed on a head halter and a body harness over the shoulders of the animal. These were monitored continuously for three days. A classification accuracy of 89.6% was achieved for the grazing, walking and resting activities (i.e., a new class combining lying and standing activity). Triaxial accelerometer data showed that sheep spent 64% (95% CI 55% to 74%) of daylight time grazing, with grazing at night reduced to 14% (95% CI 8% to 20%). Similar activity budgets were achieved from the halter mounted sensors, but not those on a body harness. These results are consistent with previous studies directly observing daily activity of pasture-based sheep and can be applied in a variety of contexts to investigate animal health and welfare metrics e.g., to better understand the impact that young sheep can suffer when carrying even modest burdens of parasitic nematodes.

Improving the Event-Based Classification Accuracy in Pit-Drilling Operations: An Application by Neural Networks and Median Filtering of the Acceleration Input Signal Data

Forestry is a complex economic sector which is relying on resource and process monitoring data. Most of the forest operations such as planting and harvesting are supported by the use of tools and machines, and their monitoring has been traditionally done by the use of pen-and-paper time studies. Nevertheless, modern data collection and analysis methods involving different kinds of platforms and machine learning techniques have been studied lately with the aim of easing the data management process. By their outcomes, improvements are still needed to reach a close to 100% activity recognition, which may depend on several factors such as the type of monitored process and the characteristics of the signals used as inputs. In this paper, we test, thought a case study on mechanized pit-drilling operations, the potential of digital signal processing techniques combined with Artificial Neural Networks (ANNs) in improving the event-based classification accuracy in the time domain. Signal processing was implemented by the means of median filtering of triaxial accelerometer data (window sizes of 3, 5, and up to 21 observations collected at 1 Hz) while the ANNs were subjected to the regularization hyperparameter’s tunning. An acceleration signal processed by a median filter with a window size of 3 observations and fed into an ANN set to learn and generalize by a regularization parameter of α = 0.01 has been found to be the best strategy in improving the event-based classification accuracy (improvements of 1% to 8% in classification accuracy depending on the type of event in question). Improvement of classification accuracy by signal filtering and ANN tuning may depend largely on the type of monitored process and its outcomes in terms of event duration; therefore, other monitoring applications may need particular designs of signal processing and ANN tuning.

OS10.5.A BrainWear: Longitudinal, objective assessment of physical activity in 42 HGG patients

Abstract BACKGROUND In patients with High Grade Glioma (HGG), QoL and physical function decline with progressive disease (PD). Objective assessment of physical functioning is challenging as patients spend most of their time away from the hospital. Wearable technology allows measurement of objective, continuous activity data in a non-obtrusive manner. BrainWear is a phase II feasibility study, collecting longitudinal physical activity (PA) data from patients with primary and secondary brain tumours. MATERIAL AND METHODS All agreed to wear an Axivity AX3 triaxial accelerometer and completed the EORTC QLQ C30 and BN20, the Montreal Cognitive Assessment (MoCA) and Multidimensional fatigue inventory (MFI) questionnaires. Accelerometers were changed at 14-day intervals, and PRO questionnaires completed at pre-specified study intervals. Age-sex matched controls were identified from the UK Biobank 7-day accelerometer study. Raw accelerometer data was processed using UK Biobank accelerometer software and inclusion of high-quality wear time selected as ≥72 hours of data in a 7-day data collection and data in each 1-hour period of a 24-hour cycle over multiple days. We analysed variation in activity by patient demographics and treatment days. The wilcoxin-signed rank test was used to compare participant activity between radiotherapy treatment days and non-treatment days, mixed effects models were used to evaluate longitudinal changes in activity and we used k-means clustering to characterise clusters of PA behaviours. RESULTS We have collected 3458 days of accelerometer data from 42 HGG patients with a median age of 59, 80% of which has been classified as high quality. Patients &amp;gt;60 years spend more time doing moderate activity compared to those &amp;lt;60 years (52 vs 33 minutes/day, p=0.012), and there are significant differences in mean vector magnitude (17.12 vs 16.85 mg, p=0.013) and walking (91 vs 72 minutes/day) between radiotherapy and non-radiotherapy days. In patients having a 6-week RT course, time spent in daily moderate activity falls 4-fold between week 1 and the second week after RT completion (70 minutes to 16 minutes/day). Comparing HGG patients to healthy controls shows a significant difference in time spent across all activities (p&amp;lt;0.05). K-means clustering analysis shows three distinct clusters, with 87% of HGG patients falling into the very inactive or moderately active groups. CONCLUSION Digital remote health monitoring is feasible and acceptable with 80% of data classified as high-quality wear-time suggesting good patient adherence. Triaxial accelerometer data collection captures objective evidence of a significant reduction in moderate daily activity at the time of expected peak RT side-effects and patients walk almost 30% less on non-RT treatment days. HGG patients show significantly lower levels of activity compared to matched healthy controls.

R package for animal behavior classification from accelerometer data-rabc.

Increasingly, animal behavior studies are enhanced through the use of accelerometry. To allow translation of raw accelerometer data to animal behaviors requires the development of classifiers. Here, we present the “rabc” (r for animal behavior classification) package to assist researchers with the interactive development of such animal behavior classifiers in a supervised classification approach. The package uses datasets consisting of accelerometer data with their corresponding animal behaviors (e.g., for triaxial accelerometer data along the x, y and z axes arranged as “x, y, z, x, y, z,…, behavior”). Using an example dataset collected on white stork (Ciconia ciconia), we illustrate the workflow of this package, including accelerometer data visualization, feature calculation, feature selection, feature visualization, extreme gradient boost model training, validation, and, finally, a demonstration of the behavior classification results.

Recurrent Neural Network for Human Activity Recognition in Embedded Systems Using PPG and Accelerometer Data

Photoplethysmography (PPG) is a common and practical technique to detect human activity and other physiological parameters and is commonly implemented in wearable devices. However, the PPG signal is often severely corrupted by motion artifacts. The aim of this paper is to address the human activity recognition (HAR) task directly on the device, implementing a recurrent neural network (RNN) in a low cost, low power microcontroller, ensuring the required performance in terms of accuracy and low complexity. To reach this goal, (i) we first develop an RNN, which integrates PPG and tri-axial accelerometer data, where these data can be used to compensate motion artifacts in PPG in order to accurately detect human activity; (ii) then, we port the RNN to an embedded device, Cloud-JAM L4, based on an STM32 microcontroller, optimizing it to maintain an accuracy of over 95% while requiring modest computational power and memory resources. The experimental results show that such a system can be effectively implemented on a constrained-resource system, allowing the design of a fully autonomous wearable embedded system for human activity recognition and logging.

US Population-referenced Percentiles for Wrist-Worn Accelerometer-derived Activity.

This study aimed to present age- and sex-specific percentiles for daily wrist-worn movement metrics in US youth and adults. This metric represents a summary of all recorded movement, regardless of the purpose, context, or intensity. Wrist-worn accelerometer data from the combined 2011-2014 National Health and Nutrition Examination Survey cycles and the 2012 National Health and Nutrition Examination Survey National Youth Fitness Survey were used for this analysis. Monitor-Independent Movement Summary units (MIMS-units) from raw triaxial accelerometer data were used. We removed the partial first and last assessment days and days with ≥5% nonwear time. Participants with ≥1 valid day were included. Mean MIMS-units were calculated across all valid days. Percentile tables and smoothed curves of daily MIMS-units were calculated for each age and sex using the Generalized Additive Models for Location Shape and Scale. The analytical sample included 14,705 participants age ≥3 yr. The MIMS-unit activity among youth was similar for both sexes, whereas adult females generally had higher MIMS-unit activity than did males. Median daily MIMS-units peaked at age 6 yr for both sexes (males, 20,613; females, 20,706). Lowest activity was observed for males and females 80+ yr of age: 8799 and 9503, respectively. Population referenced MIMS-unit percentiles for US youth and adults are a novel means of characterizing total activity volume. By using MIMS-units, we provide a standardized reference that can be applied across various wrist-worn accelerometer devices. Further work is needed to link these metrics to activity intensity categories and health outcomes.

Age Differences in Estimating Physical Activity by Wrist Accelerometry Using Machine Learning.

Accelerometer-based fitness trackers and smartwatches are proliferating with incessant attention towards health tracking. Despite their growing popularity, accurately measuring hallmark measures of physical activities has yet to be accomplished in adults of all ages. In this work, we evaluated the performance of four machine learning models: decision tree, random forest, extreme gradient boosting (XGBoost) and least absolute shrinkage and selection operator (LASSO), to estimate the hallmark measures of physical activities in young (20–50 years), middle-aged (50–70 years], and older adults (70–89 years]. Our models were built to recognize physical activity types, recognize physical activity intensities, estimate energy expenditure (EE) and recognize individual physical activities using wrist-worn tri-axial accelerometer data (33 activities per participant) from a large sample of participants (n = 253, 62% women, aged 20–89 years old). Results showed that the machine learning models were quite accurate at recognizing physical activity type and intensity and estimating energy expenditure. However, models performed less optimally when recognizing individual physical activities. F1-Scores derived from XGBoost’s models were high for sedentary (0.955–0.973), locomotion (0.942–0.964) and lifestyle (0.913–0.949) activity types with no apparent difference across age groups. Low (0.919–0.947), light (0.813–0.828) and moderate (0.846–0.875) physical activity intensities were also recognized accurately. The root mean square error range for EE was approximately 1 equivalent of resting EE [0.835–1.009 METs]. Generally, random forest and XGBoost models outperformed other models. In conclusion, machine learning models to label physical activity types, activity intensity and energy expenditure are accurate and there are minimal differences in their performance across young, middle-aged and older adults.

Reconstructing an accelerometer-based pelvis segment for three-dimensional kinematic analyses during laboratory simulated tasks with obstructed line-of-sight

Close interface between humans and inanimate objects (furniture, assistive devices, and external loads) can obstruct line-of-sight in biomechanics studies that utilize optoelectronic motion capture systems. This specific problem is frequently encountered with the pelvis segment. This study sought to compare joint and pelvis angles computed from a pelvis-fixed local coordinate system (LCS) that was constructed from optically tracked pelvis landmarks (gold standard) and landmarks derived from angular deviations calculated from triaxial accelerometer data. One participant performed seven tasks: sitting, forward bend, sit-to-stand-to-sit, forward lunge, symmetrical squat, asymmetrical squat, and gait. The root mean square error (RMSE) and coefficient of determination (R2) were examined for the pelvis, lumbar spine, and hip joint angles calculated using the standard and accelerometer-based methods for creating a LCS. The RMSE values for global pelvis angles ranged from 2.2° (gait; R2 = 0.47) to 4.9° (sit-to-stand-to-sit; R2 = 0.98), 0.6° (sitting; R2 = 0.88) to 7.4° (gait; R2 = 0.39), and 1.5° (forward bend; R2 = 0.99) to 2.9° (sit-to-stand-to-sit; R2 = 0.99) for motion about the X, Y, and Z axes, respectively. The magnitude of error observed for adjacent joint motion was lowest about the Z axis for all tasks. In conclusion, the accelerometer-based LCS offers an alternative method for computing pelvis and adjacent joint angles without the reliance on a visual line-of-sight. For motion about the X and Y axes, time-series data derived with the accelerometer-based method may be less representative of discrete events, particularly for gait and lunging tasks.

261 Accelerometer-Derived Sleep and Circadian Domains and Sociodemographic Correlates in the UK Biobank

Abstract Introduction Sleep health encompasses sleep regularity, duration, timing, efficiency and satisfaction. Accelerometry is an established method to estimate sleep in ecologically valid contexts, capturing key characteristic of rest-activity patterns, and facilitating population sleep health research. While hundreds of traits can be generated from open-source algorithms applied to raw acceleration data, the lack of clarity around their meaningful use beyond conventional measures limit the ability of these data to systematically inform evidence-based practices promoting sleep health. Here, we propose a method to identify key sleep and circadian domains, using data reduction methods for hundreds of accelerometer-derived traits to inform population-based sleep heath research. We also aimed to validate our findings by assessing whether the identified domains captured known sociodemographic associations. Methods We analyzed sociodemographic and raw triaxial accelerometer data recorded for 7 days from 79,876 adults (mean age 56.3±2.1 years, 56.3% women) participating in the UK Biobank. Standardized data processing using the open-source package GGIR (v1.7-1) resulted in the generation of 107 sleep and circadian traits. Variable clustering was used to identify key sleep and circadian domains, pertinent to sleep health, representing interpretable biological constructs minimizing correlation with other domains. Associations between identified domains and sociodemographic factors were evaluated using general linear models, and clinically significant differences were determined by standardized mean differences (SMD) ≥0.3. Results We identified 25 sleep and circadian domains explaining ≥80% of the variability of all 107 included traits. Domains capturing measures of variability tended to cluster together. The most clinically significant associations with sociodemographic characteristics were: women (vs. men) had higher sleep efficiency and lower accumulation of diurnal sleep periods; older (vs. younger) individuals had earlier most active starting time, lower acceleration amplitude and lower number of nocturnal sleep periods; and shift (vs. non-shift) workers had higher variability in sleep timing on weekends. Conclusion We demonstrate that variable clustering on accelerometer-derived data can identify meaningful sleep and circadian domains. In addition, identified domains captured known sociodemographic associations commonly observed in the sleep and circadian literature, suggesting that they could be relevant to inform public health practices that promote sleep health. Support (if any) NHLBI 5R01HL143790-02(PG); NIMHHD R01MD012734(FP); NIDA R01DA051321(FP); NIH/NHLBI K01HL123612(JM)

Evaluation of the tri-axial accelerometer to identify and predict parturition-related activities of Debouillet ewes in an intensive setting

Identifying and monitoring parturition of individual animals may help producers increase attentiveness, enabling early detection of dystocia during parturition. Parturition events are marked by subtle behavioral changes often difficult to detect by observation alone. The aim of this study was to determine the ability of tri-axial accelerometer data to accurately identify and predict parturition-related behavior of mature ewes in a pen setting. Tri-axial accelerometers recording at 12.5 Hz were placed on ear tags and attached to 13 Debouillet mature ewes before parturition. Activity was monitored 7 days prior to lambing (d −7); on the day of lambing (d 0); and 7 days post lambing (d +7). Using random forest machine learning, accelerometer data and visual observations were used to predict (i) seven mutually-exclusive behaviors; and (ii) activity (active and inactive behavior) based on five metrics calculated using variation of movements recorded by the accelerometer. The accuracy of seven predicted behaviors from an independent validation set was 66.7 %, and the accuracy for activity was 87.2 %. In addition to predicted behavior and activity, metrics calculated from accelerometer data and used for random forest predictions were evaluated 7 d before and after lambing and 12 h before and after lambing on six of the 13 ewes where the actual time of lambing was observed. No differences were detected in the seven predicted behaviors either before or after lambing. Four of five accelerometer metrics (P ≤ 0.002) were higher during the 7 d after lambing than the 7 d before lambing. Values for three of the five metrics were highest (P < 0.01) on the day of lambing. All five accelerometer metrics varied during the 12 h pre- and 12 h post parturition (P ≤ 0.004). All accelerometer metrics increased (P ≤ 0.008) 1–2 h before parturition compared to 3–4 h before parturition. In this current study, calculated direct sensor metrics served as a better indicator of lambing than predicted behaviors, processed through complex machine learning algorithms. Commercial use of accelerometers by producers may allow for detection of prolonged labor indicating potential dystocia during parturition, which may reduce lamb mortality and increase production efficiency. These results suggest that real time accelerometers could remotely monitor ewes and potentially provide managers an indication that the dam may lamb soon.

Fault Detection and Identification Method for Quadcopter Based on Airframe Vibration Signals.

Quadcopters are widely used in a variety of military and civilian mission scenarios. Real-time online detection of the abnormal state of the quadcopter is vital to the safety of aircraft. Existing data-driven fault detection methods generally usually require numerous sensors to collect data. However, quadcopter airframe space is limited. A large number of sensors cannot be loaded, meaning that it is difficult to use additional sensors to capture fault signals for quadcopters. In this paper, without additional sensors, a Fault Detection and Identification (FDI) method for quadcopter blades based on airframe vibration signals is proposed using the airborne acceleration sensor. This method integrates multi-axis data information and effectively detects and identifies quadcopter blade faults through Long and Short-Term Memory (LSTM) network models. Through flight experiments, the quadcopter triaxial accelerometer data are collected for airframe vibration signals at first. Then, the wavelet packet decomposition method is employed to extract data features, and the standard deviations of the wavelet packet coefficients are employed to form the feature vector. Finally, the LSTM-based FDI model is constructed for quadcopter blade FDI. The results show that the method can effectively detect and identify quadcopter blade faults with a better FDI performance and a higher model accuracy compared with the Back Propagation (BP) neural network-based FDI model.

Predicting the type of physical activity from tri-axial smartphone accelerometer data

Development of various statistical learning methods and their implementation in mobile device software enables moment-by-moment study of human social interactions, behavioral patterns, sleep, as well as their physical mobility and gross motor activity. Recently, through the use of supervised Machine Learning, Human Activity Recognition (HAR) has found numerous applications in biomedical engineering especially in the field of digital phenotyping. Having this in mind, in this research in order to be able to quantify the human movement activity in situ, using data from portable digital devices, we have developed code which uses Random Forest Classifier to predict the type of physical activity from tri-axial smartphone accelerometer data. The code has been written using Python programming language and Anaconda distribution of data-science packages. Raw accelerometer data was collected by using the Beiwe research platform, which is developed by the Onnela Lab at the Harvard T.H. Chan School of Public Health. Tuning has been performed by defining a grid of hyperparameter ranges, using Scikit-Learn's Randomized Search CV method, randomly sampling from the grid and performing K-Fold CV with each combination of tested values. Obtained results will enable development of more robust models for predicting the type of physical activity with more subjects, usage of different hardware, various test situations, and different environments.

Application of tri-axial accelerometer data to the interpretation of movement and behaviour of threatened black cockatoos

Abstract Context Carnaby’s (Calyptorhychus latirostris), Baudin’s (Calyptorhynchus baudinii) and forest red-tailed black cockatoos (Calyptorhynchus banksii naso) are threatened parrot species endemic to south-western Australia. Behavioural monitoring has previously involved direct observation, which has proven challenging because of their cryptic nature, the type of habitat they move through and their speed of movement. The development of a model to accurately classify behaviour from tri-axial accelerometer data will provide greater insight into black cockatoo behaviour and ecology. Aims To develop an automated classifier model to classify accelerometer data from released black cockatoos to determine behaviour and activity budgets for three species of black cockatoo. Methods In the present study, we attached tri-axial accelerometers, housed in GPS tags, to four Carnaby’s cockatoos, three forest red-tailed black cockatoos and two Baudin’s cockatoos in captive care, undergoing rehabilitation for release back to the wild. Accelerometer data from these birds was coupled with 19 video files of the birds’ behaviour when flying, feeding and resting, to develop an automated behaviour classifier. The classifier was then used to annotate accelerometer data from 15 birds released after successful rehabilitation and to calculate activity budgets for these birds post-release. Key results We developed a classifier able to identify resting, flying and foraging behaviours from accelerometer data with 86% accuracy, as determined by the percentage of observed behaviours correctly identified by the classifier. The application of the classifier to accelerometer data from 15 released cockatoos enabled us to determine behaviours and activity budgets for all three species of black cockatoo. Black cockatoos spent most of their time at rest, followed by foraging with a short period of time flying. Conclusions Application of the classifier to data from released birds gives researchers the ability to remotely identify patterns of behaviour and calculate activity budgets. Implications Combining behaviour and activity budgets with location data provides useful insight into cockatoo movement, distribution, and habitat use. Such information is important for informing conservation efforts and addressing outstanding research objectives. Further studies including larger sample sizes of Baudin’s and forest red-tailed black cockatoos and comparing behaviour and activity between birds in breeding and non-breeding areas are warranted.

Characteristic of Northern Thailand Seismic Stations from Horizontal-to-Vertical Spectral Ratios (HVSR) Analysis

Several major cities in Northern Thailand are located on a sedimentary basin, which could amplify ground motion from future earthquakes as they lie along several active faults. To study the seismic response and natural periods of these basins, a Horizontal to Vertical Spectral Ratio (HVSR) for each seismic station in Northern Thailand has been proposed. Several observed ground motion from moderate to small seismic events in and around Northern Thailand were used in this analysis. Several seismometers (TSA-100S, Trillium-120, PA-23, S-13 and KS 2000) of the Department of Mineral Resources (DMR) and Thai Meteorological Department (TMD) with triaxial digital accelerometer data were used. The 3 axis digital accelerometers (2-horizontal and vertical) were converted to Fourier amplitude spectral (FAS) in order to find basin natural period by Horizontal to Vertical Spectral Ratio (HVSR). This research investigates 6 stations in Northern Thailand with at least 10 observed ground motions for each station. We found that the natural period of three stations (CRAI, PAYA and PHRA) showed strong site amplification effect at low natural period (between 0.1 - 0.2 s with an amplitude larger than 2). The natural period of CMMT and MHIT stations were insignificant since CMMT and MHIT were located on the rocks. Lastly, we observed long natural period for CMCA station located in Chiang Mai basin around 2.50 s with an amplitude larger than 3. Furthermore, the computed site natural periods were compared with the average top 30 m shear wave velocity (Vs30) based on the data from the Shuttle Radar Topography Mission (SRTM) and Multichannel Analysis of Surface Wave (MASW) to find the type of soil in each area.

Real-Time Physical Activity Recognition on Smart Mobile Devices Using Convolutional Neural Networks

Given the ubiquity of mobile devices, understanding the context of human activity with non-intrusive solutions is of great value. A novel deep neural network model is proposed, which combines feature extraction and convolutional layers, able to recognize human physical activity in real-time from tri-axial accelerometer data when run on a mobile device. It uses a two-layer convolutional neural network to extract local features, which are combined with 40 statistical features and are fed to a fully-connected layer. It improves the classification performance, while it takes up 5–8 times less storage space and outputs more than double the throughput of the current state-of-the-art user-independent implementation on the Wireless Sensor Data Mining (WISDM) dataset. It achieves 94.18% classification accuracy on a 10-fold user-independent cross-validation of the WISDM dataset. The model is further tested on the Actitracker dataset, achieving 79.12% accuracy, while the size and throughput of the model are evaluated on a mobile device.