Mean Root Mean Square Error Research Articles

Comparison of different machine learning approaches for tropospheric profiling based on COSMIC-2 data

Precise and reliable information on the tropospheric temperature and water vapor profiles play a key role in weather and climate studies. Among the sensors supporting the observations of the troposphere, one can distinguish the Global Navigation Satellite System Radio Occultation (RO) technique, which provides accurate and high-quality meteorological profiles. However, external knowledge about temperature is essential to estimate other physical atmospheric parameters. To overcome this constraint, I trained and evaluated four different machine learning models comprising Artificial Neural Network (ANN) and Random Forest regression algorithms, where no auxiliary meteorological data is needed. To develop the models, I employed 150,000 globally distributed (45°S–45°N) RO profiles between October 2019 and December 2020. Input vectors consisted of bending angle or refractivity profiles from the Formosa Satellite-7/Constellation Observing System for Meteorology, Ionosphere, and Climate-2 mission together with the month, hour, and latitude of the RO event. While temperature, pressure, and water vapor profiles derived from the modern ERA5 reanalysis and interpolated to the RO location served as the models’ targets. Evaluation on the testing data set revealed a good agreement between all model outputs and ERA5 targets, where slightly better statistics were noted for ANN and refractivity inputs. Vertically averaged root mean square error (RMSE) did not exceed 1.7 K for the temperature and reached around 1.4 hPa and 0.45 hPa for the total and water vapor pressures. Additional validation with 477 co-located radiosonde observations and the operational one-dimensional variational product showed slightly larger discrepancies with the mean RMSE of around 1.9 K, 1.9 hPa, and 0.5 hPa for the temperature, pressure, and water vapor, respectively.Graphical

Machine Learning-Based Estimation of Ground Reaction Forces and Knee Joint Kinetics from Inertial Sensors While Performing a Vertical Drop Jump.

Nowadays, the use of wearable inertial-based systems together with machine learning methods opens new pathways to assess athletes’ performance. In this paper, we developed a neural network-based approach for the estimation of the Ground Reaction Forces (GRFs) and the three-dimensional knee joint moments during the first landing phase of the Vertical Drop Jump. Data were simultaneously recorded from three commercial inertial units and an optoelectronic system during the execution of 112 jumps performed by 11 healthy participants. Data were processed and sorted to obtain a time-matched dataset, and a non-linear autoregressive with external input neural network was implemented in Matlab. The network was trained through a train-test split technique, and performance was evaluated in terms of Root Mean Square Error (RMSE). The network was able to estimate the time course of GRFs and joint moments with a mean RMSE of 0.02 N/kg and 0.04 N·m/kg, respectively. Despite the comparatively restricted data set and slight boundary errors, the results supported the use of the developed method to estimate joint kinetics, opening a new perspective for the development of an in-field analysis method.

Nonlinear ARIMA Models with Feedback SVR in Financial Market Forecasting

In recent years, as global financial markets have become increasingly connected, the degree of correlation between financial assets has become closer, and technological advances have made the transmission of information faster and faster, and information networks have integrated capital markets into one, making it easier for single financial market risk problems to form systemic risk through a high degree of market linkage effects. Based on the characteristics of financial markets containing both linear and nonlinear components, this paper chooses to use Autoregressive Integrated Moving Average (ARIMA) model and feedback Support Vector Regression (SVR) models to effectively integrate the ARIMA model and the SVR model, taking into account their respective linear and nonlinear characteristics. The paper chooses to use the (Autoregressive Integrated Moving Average (ARIMA) model and feedback Support Vector Regression (SVR) models to effectively integrate the strengths of the ARIMA and SVR models in terms of linearity and nonlinearity to perform forecasting analysis of financial markets. One of the important functions of forecasting is to transform future uncertainty into measurable risk, so that we can base our plans and actions on it. In this paper, the combined ARIMA-SVR model is compared with the single ARIMA model and SVR model in terms of the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE), where MAE and RMSE measure the absolute error between the predicted and true values, and MAPE measures the relative error between the predicted and true values. and the relative error between the true value. The results show that the combined ARIMA-SVR model has a better forecasting effect and higher forecasting accuracy than the single ARIMA model and SVR model, and the SVR model has higher forecasting accuracy than the ARIMA model in forecasting financial markets.

A functional modelling approach for reconstructing 3 and 4 dimensional wet refractivity fields in the lower atmosphere using GNSS measurements

• 3D and 4D functional model-based atmosphere tomography have been developed. • Spherical Cap harmonics formed the horizontal basis for the functional model. • Empirical orthogonal functions formed the vertical basis of the functional model. • Spline basis function has been used to develop a time correlated functional model. • The results show higher accuracy of implemented 4D tomography method. The GNSS tomography provides the opportunity to estimate atmospheric wet refractivity, which is important for precise positioning and navigation, as well as for constraining weather and climate models. The GNSS-derived water vapor is often estimated by implementing voxel-based inversions, but this technique is numerically unstable due to the high number of unknown parameters especially in networks covering large areas or when the spatial resolution increases. To mitigate this problem, in this study, we introduce functional based 3 and 4 dimensional (3D and 4D) inversion formulations. Here, the horizontal changes are modelled by the spherical cap harmonic functions. For the vertical component, empirical orthogonal functions derived from ERA5 (Empirical Reanalysis Fifth generation) are applied as background model. The time-dependency is accounted for by applying polynomial spline functions. Numerical results are based on a network of about 190 GPS stations in Germany during 15 days in summer and winter of 2018. Observations from 8 radiosonde stations are applied for comparisons. Our results indicate that the functional tomography is effective and retrieves wet refractivity indices with the mean RMSE (Root Mean Square Error) of about 1.9 ppm. These values are found to be up to 22% smaller than those derived by comparing ERA5 with the radiosonde data.

Situational Awareness: The Effect of Stimulus Type and Hearing Protection on Sound Localization.

The purpose of the current study was to test sound localization of a spoken word, rarely studied in the context of localization, compared to pink noise and a gunshot, while taking into account the source position and the effect of different hearing protection devices (HPDs) used by the listener. Ninety participants were divided into three groups using different HPDs. Participants were tested twice, under with- and no-HPD conditions, and were requested to localize the different stimuli that were delivered from one of eight speakers evenly distributed around them (starting from 22.5°). Localization of the word stimulus was more difficult than that of the other stimuli. HPD usage resulted in a larger mean root-mean-square error (RMSE) and increased mirror image reversal errors for all stimuli. In addition, HPD usage increased the mean RMSE and mirror image reversal errors for stimuli delivered from the front and back, more than for stimuli delivered from the left and right. HPDs affect localization, both due to attenuation and to limitation of pinnae cues when using earmuffs. Difficulty localizing the spoken word should be considered when assessing auditory functionality and should be further investigated to include HPDs with different attenuation spectra and levels, and to further types of speech stimuli.

Assessing inclination angles of tree branches from terrestrial laser scan data using a skeleton extraction method

Assessing the inclination angles (IAs) of tree branches is essential to evaluate their rainfall interception ability. However, depicting the IA distribution within a forest canopy is still a challenge. This study developed the WoodSKE method to extract skeletons from discrete point clouds of tree branches collected by terrestrial laser scanners (TLSs) for assessing their IAs. A TLS point cloud of tree branches was firstly contracted according to the pointwise local point distribution pattern to extract its coarse skeleton. Then, the coarse skeleton would be thinned and optimized by a noise filtering method. The IAs of tree branches were estimated based on their skeleton distribution. For each point cloud that has the reference skeleton, the average and root mean squared error (RMSE) of offset distance for its WoodSKE-extracted skeleton was less than 0.011 m and less than 0.019 m, respectively. Furthermore, the WoodSKE method was robust to process point clouds with noise. Comparing to the measured IAs at sample points selected from tested point clouds, the mean absolute error and RMSE of their skeleton-assessed IAs were 7˚ and 11.7˚, respectively. Over 86% of the number of sample points in each tested point cloud with IA assessment error lower than 15˚. Extracting skeletons from TLS point clouds of tree branches provides a base for assessing the distribution of IA-related structure features within a forest canopy.

Comparing Four Types Methods for Karst NDVI Prediction Based on Machine Learning

As a link for energy transfer between the land and atmosphere in the terrestrial ecosystem, karst vegetation plays an important role. Karst vegetation is not only affected by environmental factors but also by intense human activities. The nonlinear characteristics of vegetation growth are induced by the interaction mechanism of these factors. Previous studies of this relationship were not comprehensive, and it is necessary to further explore it using a suitable method. In this study, we selected climate, human activities, topography, and soil texture as the response factors; a nonlinear relationship model between the karst normalized difference vegetation index (NDVI) and these factors was established by applying a back propagation neural network (BPNN), a radial basis function neural network (RBFNN), the random forest (RF) algorithm, and support vector regression (SVR); and then, the karst NDVI was predicted. The coefficient of determination (R2), mean square error (MSE), root mean square error (RMSE), and mean absolute percentage error (MAPE) of the obtained results were calculated, and the mean R2 values of the BPNN, RBFNN, RF, and SVR models were determined to be 0.77, 0.86, 0.89, and 0.91, respectively. Compared with the BPNN, RBFNN, and RF models, the SVR model had the lowest errors, with mean MSE, RMSE, and MAPE values of 0.001, 0.02, and 2.77, respectively. The results show that the BPNN, RBFNN, RF, and SVR models are within acceptable ranges for karst NDVI prediction, but the overall performance of the SVR model is the best, and it is more suitable for karst vegetation prediction.

A targeted solution for estimating the cell-type composition of bulk samples

BackgroundTo avoid false-positive findings and detect cell-type specific associations in methylation and transcription investigations with bulk samples, it is critical to know the proportions of the major cell-types.ResultsWe present a novel approach that allows for precise estimation of cell-type proportions using only a few highly informative methylation markers. The most reliable estimates were obtained with 17 amplicons (34 CpGs) using the MuSiC estimator, for which the average correlations between the estimated and the true cell-type proportions were 0.889. Furthermore, the estimates were not significantly different from the true values (P = 0.95) indicating that the estimator is unbiased and the standard deviation of the estimates further indicate high precision. Moreover, the overall variability of the estimates as measured by the Root Mean Squared Error (RMSE), which is a function of both bias and precision, was low (mean RMSE = 0.038). Taken together, these results indicate that the approach produced reliable estimates that are both unbiased and highly precise.ConclusionThis cost-effective approach for estimating cell-type proportions in bulk samples allows for enhanced targeted analysis, which in turn will minimize the risk of reporting false-positive findings and allowing for detection of cell-type specific associations. The approach is applicable across platforms and can be extended to assess cell-type proportions for various tissues.

Real-time CT image generation based on voxel-by-voxel modeling of internal deformation by utilizing the displacement of fiducial markers.

To show the feasibility of real-time CT image generation technique utilizing internal fiducial markers that facilitate the evaluation of internal deformation. In the proposed method, a linear regression model that can derive internal deformation from the displacement of fiducial markers is built for each voxel in the training process before the treatment session. Marker displacement and internal deformation are derived from the four-dimensional computed tomography (4DCT) dataset. In the treatment session, the three-dimensional deformation vector field is derived according to the marker displacement, which is monitored by the real-time imaging system. The whole CT image can be synthesized by deforming the reference CT image with a deformation vector field in real-time. To show the feasibility of the technique, image synthesis accuracy and tumor localization accuracy were evaluated using the dataset generated by extended NURBS-Based Cardiac-Torso (XCAT) phantom and clinical 4DCT datasets from six patients, containing 10 CT datasets each. In the validation with XCAT phantom, motion range of the tumor in training data and validation data were about 10 and 15mm, respectively, so as to simulate motion variation between 4DCT acquisition and treatment session. In the validation with patient 4DCT dataset, eight CT datasets from the 4DCT dataset were used in the training process. Two excluded inhale CT datasets can be regarded as the datasets with large deformations more than training dataset. CT images were generated for each respiratory phase using the corresponding marker displacement. Root mean squared error (RMSE), normalized RMSE (NRMSE), and structural similarity index measure (SSIM) between the original CT images and the synthesized CT images were evaluated as the quantitative indices of the accuracy of image synthesis. The accuracy of tumor localization was also evaluated. In the validation with XCAT phantom, the mean NRMSE, SSIM, and three-dimensional tumor localization error were 7.5±1.1%, 0.95±0.02, and 0.4±0.3mm, respectively. In the validation with patient 4DCT dataset, the mean RMSE, NRMSE, SSIM, and three-dimensional tumor localization error in six patients were 73.7±19.6 HU, 9.2±2.6%, 0.88±0.04, and 0.8±0.6mm, respectively. These results suggest that the accuracy of the proposed technique is adequate when the respiratory motion is within the range of the training dataset. In the evaluation with a marker displacement larger than that of the training dataset, the mean RMSE, NRMSE, and tumor localization error were about 100 HU, 13%, and <2.0mm, respectively, except for one case having large motion variation. The performance of the proposed method was similar to those of previous studies. Processing time to generate the volumetric image was <100ms. We have shown the feasibility of the real-time CT image generation technique for volumetric imaging.

A Comparative Analysis of UAV-RTK and UAV-PPK Methods in Mapping Different Surface Types

This study aimed to compare unmanned aerial vehicle (UAV) based real-time kinematic (RTK) and post-processing kinematic (PPK) methods via five approaches: an RTK-CORS method (M1), a short-baseline PPK method obtaining corrections from a GNSS base station (M2), and three long-baseline PPK methods that obtained corrections from the three Turkish RTK-CORS network TUSAGA-Aktif reference stations (M3: IZMI, M4: CESM, and M5: KIKA). The comparison was based on the accuracy of the corrected camera positions, the average error of the camera locations computed in the photo-alignment and optimization process, georeferencing errors of the models via nine GCPs based on four scenarios, and Root Mean Square (RMS) errors in the Z-direction for different surface types (i.e. roads, shadows, shrubs, boulders, trees, and ground). For the surface types of “ground”, “roads”, and “shrubs”, RMS error rates were obtained 10 cm lower than that of other surface types in all methods except M4. The greatest differences were obtained over trees and shadowed areas. The conclusion of these comparisons was that the lowest RMS error rate was determined on a solid textured surface. The consideration of mean RMS error regardless of surface type in such model comparisons is misleading.

Real-time Prediction of the Daily Incidence of COVID-19 in 215 Countries and Territories Using Machine Learning: Model Development and Validation.

BackgroundAdvanced prediction of the daily incidence of COVID-19 can aid policy making on the prevention of disease spread, which can profoundly affect people's livelihood. In previous studies, predictions were investigated for single or several countries and territories.ObjectiveWe aimed to develop models that can be applied for real-time prediction of COVID-19 activity in all individual countries and territories worldwide.MethodsData of the previous daily incidence and infoveillance data (search volume data via Google Trends) from 215 individual countries and territories were collected. A random forest regression algorithm was used to train models to predict the daily new confirmed cases 7 days ahead. Several methods were used to optimize the models, including clustering the countries and territories, selecting features according to the importance scores, performing multiple-step forecasting, and upgrading the models at regular intervals. The performance of the models was assessed using the mean absolute error (MAE), root mean square error (RMSE), Pearson correlation coefficient, and Spearman correlation coefficient.ResultsOur models can accurately predict the daily new confirmed cases of COVID-19 in most countries and territories. Of the 215 countries and territories under study, 198 (92.1%) had MAEs <10 and 187 (87.0%) had Pearson correlation coefficients >0.8. For the 215 countries and territories, the mean MAE was 5.42 (range 0.26-15.32), the mean RMSE was 9.27 (range 1.81-24.40), the mean Pearson correlation coefficient was 0.89 (range 0.08-0.99), and the mean Spearman correlation coefficient was 0.84 (range 0.2-1.00).ConclusionsBy integrating previous incidence and Google Trends data, our machine learning algorithm was able to predict the incidence of COVID-19 in most individual countries and territories accurately 7 days ahead.

Assessment of Five Thermal Infrared Kernel-Driven Models Using Limited Multiangle Observations

There are five widely used kernel-driven models in the thermal infrared domain designed for the angular correction of land surface temperature (LST), including three-parameter Roujean Lagouarde (RL), Vinnikov, RossThick-LiSparseR (Ross–Li), LiStrahlerFriedl-LiDenseR (LSF-Li), and four-parameter Vinnikov-RoujeanLagouarde (Vinnikov-RL). Their fitting accuracies with hundreds of observation angles (i.e., sufficient angle) were studied; however, the fitting ability of these five models with limited observation angles is unknown, which makes it difficult to choose the appropriate one in applications. To solve this problem, 30 600 groups of multiangle directional brightness temperature (DBT) datasets were simulated by the unified optical-thermal 4-stream model considering scattering by arbitrary inclined leaves (4SAIL) model considering ten different leaf area index values, three leaf inclination distribution functions, two hotspot factors, 17 different component temperatures, five solar zenith angles, and six solar azimuth angles. Each group contains DBT values in 21 960 viewing directions [i.e., 61 viewing zenith angle (VZA) <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times $ </tex-math></inline-formula> 360 viewing azimuth angle (VAA)]. We assume that all limited observations are in the plane with VAA = 180°/0° and VZA changing from −60° to 60° with a step of 10°. There are 13 candidate angles to be selected. Five, seven, nine, and 11 angle sampling schemes include 225, 400, 225, and 36 limited multiangle combinations, respectively. Each combination was used to drive these five kernel-driven models to fit 21 960 DBTs for 30 600 groups of 4SAIL simulations. The root-mean-square error (RMSE) of each combination and mean RMSE of all 886 combinations were used to assess the overall fitting ability of five kernel-driven models. In addition, 1 k errors were added to the driven DBTs to evaluate the models’ robustness. Four groups of airborne measured DBTs were adopted to validate the assessment conclusions. Results show that the recommended order of these five models driven by 5–11 multiangle DBTs is Vinnikov-RL, LSF-Li, Vinnikov, Ross–Li, and RL when the driven DBTs do not contain errors; Vinnikov-RL, Vinnikov, LSF-Li, Ross–Li, and RL when the driven DBTs contain 1k errors; and Vinnikov-RL, LSF-Li, Ross–Li, RL, and Vinnikov for four groups of airborne measured datasets.

Prediction of postoperative refractive astigmatism before toric intraocular lens implantation

BackgroundTo determine the preoperative factors influencing refractive astigmatism after cataract surgery for astigmatism correction by toric intraocular lens (IOL) implantation and to evaluate the prediction model using these factors.MethodsProspective, observational case series. The right eyes of forty consecutive patients with preoperative corneal astigmatism of the total cornea of 1.5 diopters (D) or more in magnitude and scheduled for implantation of a non-toric IOL during cataract surgery with a 2.4-mm temporal clear corneal incision were examined prospectively. The vertical/horizontal astigmatism component (J0) and oblique astigmatism component (J45) of refractive and corneal astigmatism were converted using power vector analysis. Multivariate regression analysis was performed with refractive astigmatism at three months postoperatively as the dependent variable, and preoperative parameters including age, sex, refractive astigmatism, corneal astigmatism, sphere, spherical equivalent, intraocular pressure, corneal thickness, anterior chamber depth, lens thickness, lens positions (tilt and decentration), axial length, and corneal higher order aberrations as independent variables. The root mean square (RMS) errors were calculated to express the regression model fit.ResultsThe regression model for the J0 component was Postoperativekern0.34em refractivekern0.2em J0=1.05times Conealkern0.2em J0-0.14 (R2 = 0.96, P < 0.001). The model for the J45 component was Postoperativekern0.34em refractivekern0.2em J45=0.68times Conealkern0.2em J45+0.19times Preoperativekern0.34em refractivekern0.2em J45-0.06 (R2 = 0.72, P < 0.001). The mean RMS errors for preoperative corneal astigmatism alone and the multivariate model were 0.58 D and 0.46 D, respectively. There was a statistically significant difference between them (P = 0.02).ConclusionsRefractive astigmatism after implantation of a toric IOL can be predicted by the regression model more accurately than by corneal astigmatism alone. However, the prediction of oblique astigmatism remains a challenge.

The Penetration Analysis of Airborne Ku-Band Radar Versus Satellite Infrared Lidar Based on the Height and Energy Percentiles in the Boreal Forest

The paper investigates the penetration properties of an airborne Ku-band frequency modulated continuous waveform (FMCW) profiling radar named Tomoradar and a satellite near-infrared lidar into the boreal forest of Finland. We achieve the accumulative energy distributions based on the Tomoradar waveforms and the satellite lidar waveforms generated from the high-density airborne lidar data within Tomoradar footprints. By comparing two groups of the height percentiles and energy percentiles derived from the accumulative energy distributions, we evaluate the relationship of penetrations between the Ku-band microwave and near-infrared laser according to the coefficients of the determination (COD), and the root mean square errors (RMSE) of linear regression analyses. The quantitative analysis results demonstrate that the height and energy percentiles derived from Tomoradar waveforms correlate well with those from satellite lidar waveforms with the mean correlation coefficients of more than 0.78 and 0.85. The linear regression models for the height and energy percentile produce excellent fits with the mean CODs of 0.95 and 0.90 and the mean RMSEs of 1.25 m and 0.03, respectively. Less than 15% of height percentiles and 87.54% of the energy percentiles in the sixth stratum near the ground derived from Tomoradar waveforms surpass those from satellite lidar waveforms. Hence, the Ku-band microwave can penetrate deeper into the forest than the near-infrared laser at the same spatial scale. In addition, quadratic fitting models are established to describe the differences of the height percentile (DHP) and the energy percentile (DEP) to expound the canopy height and closure contributions numerically. The facts that the CODs of the DHP and DEP individually are more than 0.96 and 0.89 and the fitting residual histograms approximate to normal distributions reveal the reliabilities of the proposed fitting models. Thus, the penetration analyses are valid for the explorations on the FMCW radar applications and the data fusion of the Ku-band radar and near-infrared lidar in the forest investigations.

A Reduced-Order Computational Model of a Semi-Active Variable-Stiffness Foot Prosthesis.

Passive energy storage and return (ESR) feet are current performance standard in lower limb prostheses. A recently developed semi-active variable-stiffness foot (VSF) prosthesis balances the simplicity of a passive ESR device with the adaptability of a powered design. The purpose of this study was to model and simulate the ESR properties of the VSF prosthesis. The ESR properties of the VSF were modeled as a lumped parameter overhung beam. The overhung length is variable, allowing the model to exhibit variable ESR stiffness. Foot-ground contact was modeled using sphere-to-plane contact models. Contact parameters were optimized to represent the geometry and dynamics of the VSF and its foam base. Static compression tests and gait were simulated. Simulation outcomes were compared to corresponding experimental data. Stiffness of the model matched that of the physical VSF (R2: 0.98, root-mean-squared error (RMSE): 1.37 N/mm). Model-predicted resultant ground reaction force (GRFR) matched well under optimized parameter conditions (R2: 0.98, RMSE: 5.3% body weight,) and unoptimized parameter conditions (R2: 0.90, mean RMSE: 13% body weight). Anterior-posterior center of pressure matched well with R2 > 0.94 and RMSE < 9.5% foot length in all conditions. The ESR properties of the VSF were accurately simulated under benchtop testing and dynamic gait conditions. These methods may be useful for predicting GRFR arising from gait with novel prostheses. Such data are useful to optimize prosthesis design parameters on a user-specific basis.

Air temperature prediction using different machine learning models

Air temperature is an essential climatic component particularly in water resources management and other agro-hydrological/meteorological activities planning This paper examines the prediction capability of three machine learning models, least square support vector machine (LSSVM), group method and data handling neural network (GMDHNN) and classification and regression trees (CART) in air temperature forecasting using monthly temperature data of Astore and Gilgit climatic stations of Pakistan. The prediction capability of three machine learning models is evaluated using different time lags input combinations with help of root mean square error (RMSE), the mean absolute error (MAE) and coefficient of determination (R&lt;sup&gt;2&lt;/sup&gt;).statistical indicators. The obtained results indicated that the LSSVM model is more accurate in temperature forecasting than GMDHNN and CART models. LSSVM significantly decreases the mean RMSE of the GMHNN and CART models by 1.47-3.12% and 20.01-25.12% for the Chakdara and Kalam Stations, respectively.

Four-dimensional inverse-geometry computed tomography: a preliminary study

This study introduces and evaluates respiratory-correlated four-dimensional (4D) inverse geometry computed tomography (IGCT). The projection data of the IGCT were acquired in a single gantry rotation over 120 s. Three virtual phantoms—static Defrise, 4D Shepp–Logan, and 4D extended cardiac-torso (XCAT)—were used to obtain projection data for the IGCT and cone-beam computed tomography (CBCT). The projection acquisition parameters were determined to eliminate vacancies in the Radon space for an accurate rebinning process. Phase-based sorting was conducted within 10 phase bins, and the sorted projection data were binned into a cone beam geometry. Finally, Feldkamp–Davis–Kress reconstruction was conducted independently at each phase. The reconstructed images were compared using the structural similarity index measure (SSIM) and root mean square error (RMSE). The vertical profile of the Defrise reconstruction image was uniform, and the cone beam artefact was reduced in the IGCT image. Under an ideal projection acquisition condition, the mean coronal plane SSIMs of the Shepp–Logan and 4D XCAT phantoms were 0.899 and 0.706, respectively, which were higher than those of the CBCT (0.784 and 0.623, respectively). Similarly, the mean RMSEs of the coronal plane IGCT (0.036 and 0.158) exhibited an improvement over those of the CBCT (0.165 and 0.261, respectively). The mean standard deviations of the SSIM and RMSE were lower for IGCT than for CBCT. In particular, the SSIM and RMSE of the sagittal and coronal planes of the Shepp–Logan IGCT images were stable in all phase bins; however, those of the CBCT changed depending on the phase bins. Poor image quality was observed for IGCT under inappropriate conditions. This was caused by a vacancy in the Radon space, owing to an inappropriate scan setting. Overall, the proposed 4D IGCT exhibited better image quality than conventional CBCT.

A Regression Method With Subnetwork Neurons for Vigilance Estimation Using EOG and EEG

In recent years, it has been observed that there is an increasing rate of road accidents due to the low vigilance of drivers. Thus, the estimation of drivers' vigilance state plays a significant role in public transportation safety. We have adopted a feature fusion strategy that combines the electroencephalogram (EEG) signals collected from various sites of the human brain, including forehead, temporal, and posterior and forehead electrooculogram (forehead-EOG) signals, to address this factor. The level of vigilance is predicted through a new learning model known as double-layered neural network with subnetwork nodes (DNNSNs), which comprises several subnetwork nodes, and each node in turn is composed of many hidden nodes that have various capabilities of feature selection (dimension reduced), feature learning, etc. The proposed single modality that uses only forehead-EOG signal exhibits a mean root-mean-square error (RMSE) of 0.12 and a mean Pearson product-moment correlation coefficient (COR) of 0.78. On one hand, an EEG signal achieved a mean RMSE of 0.13 and a mean COR of 0.72. Whereas, on the other, the proposed multimodality achieved values of 0.09 and 0.85 for the mean RMSE and the mean COR, respectively. Experimental results show that the proposed DNNSN with multimodality fusion outperforms the model with single modality for vigilance estimation due to the complementary information between forehead-EOG and EEG. After a favorable learning rate was applied to the input layer, the mean RMSE/COR improved to 0.11/0.79, 0.12/0.74, and 0.08/0.86, respectively. Hence, this quantitative analysis proves that the proposed method provides better feasibility and efficiency learning capability and surmounts other state-of-the-art techniques.

Fractional monitoring of desert vegetation degradation, recovery, and greening using optimized multi-endmembers spectral mixture analysis in a dryland basin of Northwest China

ABSTRACT Accurate identification of desert vegetation dynamics in arid regions is challenging because of its complex composition of grass species, obscure boundary with non-desert vegetation, and high sensitiveness to climatic variations. This study examined the ability of optimized multi-endmember spectral mixture analysis (MESMA) for monitoring desert vegetation degradation, recovery, and greening in a dryland basin of Northwest China using Landsat time series data from the 1990s to 2016. Eight modeled endmember fractions were generated using the best endmember model with the lowest fraction error and root mean square error (RMSE). Abundances of non-desert vegetation, desert vegetation, soil, and impervious surface areas were incorporated based on the eight original fractions and validated using high spatial resolution images. Finally, the post-classification comparison approach was used to detect desert vegetation degradation, recovery, and greening. Results show that: (1) More than 97% of the land pixels were modeled successfully into eight endmember fractions for each period with the mean RMSE less than 0.01. All four simulated abundances had high correlations (r = 0.89–0.96) with the corresponding reference data, indicating good performance of MESMA in this study; (2) Desert vegetation increased dramatically (772.68 km2) during the 26-year period. The major change was desert vegetation recovery with a total area of 10,705 km2, followed by degradation with a total area of 4,715 km2. The greening area was the smallest, covering only 1,509 km2; (3) Increased precipitation was the major contributor for desert vegetation greening in the west of upper region while decreased precipitation was the major contributor for the degradation in the west of lower region. Anthropogenic factors (e.g., improvement of irrigation, crop expansion) were major contributors for the change in desert vegetation in the middle region. This research demonstrates that MESMA is promising in detecting desert vegetation dynamics in semi-arid and arid regions.

Development, Calibration and Validation of an Internal Air Temperature Model for a Naturally Ventilated Nearly Zero Energy Building: Comparison of Model Types and Calibration Methods

In this study, a grey box (GB) model for simulating internal air temperatures in a naturally ventilated nearly zero energy building (nZEB) was developed and calibrated, using multiple data configurations for model parameter selection and an automatic calibration algorithm. The GB model was compared to a white box (WB) model for the same application using identical calibration and validation datasets. Calibrating the GB model using only one week of data produced very accurate results for the calibration periods but led to inconsistent and typically inaccurate results for the validation periods (root mean squared error (RMSE) in validation periods was 229% larger than the RMSE in calibration periods). Using three weeks of data from varying seasons for calibration reduced the model accuracy in the calibration period but substantially increased the model accuracy and generalisation abilities for the validation period, reducing the mean RMSE by over 160%. The use of one week of data increased the standard deviation in parameter selections by over 40% when compared with the three-week calibration datasets. Utilising data from multiple seasons for calibration purposes was found to substantially improve generalisation abilities. When compared to the WB model, the GB model produced slightly less accurate results (mean RMSE of the GB model was 1.5% higher). However, the authors found that employing GB modelling with an automatic model calibration technique reduced the human labour input for simulating internal air temperature of a naturally ventilated nZEB by approximately 90%, relative to WB modelling using a manually calibrated approach.