Search

Deep learning-based breast lesion detection in ultrasound images has demonstrated great potential to provide objective suggestions for radiologists and improve their accuracy in diagnosing breast diseases. However, the lack of an effective feature enhancement approach limits the performance of deep learning models. Therefore, in this study, we propose a novel dual global attention neural network (DGANet) to improve the accuracy of breast lesion detection in ultrasound images. Specifically, we designed a bilateral spatial attention module and a global channel attention module to enhance features in spatial and channel dimensions, respectively. The bilateral spatial attention module enhances features by capturing supporting information in regions neighboring breast lesions and reducing integration of noise signal. The global channel attention module enhances features of important channels by weighted calculation, where the weights are decided by the learned interdependencies among all channels. To verify the performance of the DGANet, we conduct breast lesion detection experiments on our collected data set of 7040 ultrasound images and a public data set of breast ultrasound images. YOLOv3, RetinaNet, Faster R-CNN, YOLOv5, and YOLOX are used as comparison models. The results indicate that DGANet outperforms the comparison methods by 0.2%-5.9% in total mean average precision. [ABSTRACT FROM AUTHOR]

• A systematic overview on integrating medical domain knowledge into deep models. • Different kinds of domain knowledge and their integrating methods are summarized. • Challenges and future directions of integrating domain knowledge are discussed. Although deep learning models like CNNs have achieved great success in medical image analysis, the small size of medical datasets remains a major bottleneck in this area. To address this problem, researchers have started looking for external information beyond current available medical datasets. Traditional approaches generally leverage the information from natural images via transfer learning. More recent works utilize the domain knowledge from medical doctors, to create networks that resemble how medical doctors are trained, mimic their diagnostic patterns, or focus on the features or areas they pay particular attention to. In this survey, we summarize the current progress on integrating medical domain knowledge into deep learning models for various tasks, such as disease diagnosis, lesion, organ and abnormality detection, lesion and organ segmentation. For each task, we systematically categorize different kinds of medical domain knowledge that have been utilized and their corresponding integrating methods. We also provide current challenges and directions for future research. [ABSTRACT FROM AUTHOR]

The visual storytelling (VIST) task aims at generating reasonable, human-like and coherent stories with the image streams as input. Although many deep learning models have achieved promising results, most of them do not directly leverage the sentiment information of stories. In this paper, we propose a sentiment-aware generative model for VIST called SentiStory. The key of SentiStory is a multi-layered sentiment extraction module (MLSEM). For a given image stream, the higher layer gives coarse-grained but accurate sentiments, while the lower layer of the MLSEM extracts fine-grained but usually unreliable ones. The two layers are combined strategically to generate coherent and rich visual sentiment concepts for the VIST task. Results from both automatic and human evaluations demonstrate that with the help of the MLSEM, SentiStory achieves improvement in generating more coherent and human-like stories. [ABSTRACT FROM AUTHOR]

• Introverts are susceptible to music, and prefer slow tempo and classical music. • Pop music aroused more than classical one and induce closer headway time. • Medium music tempo was most appropriate for keeping stable car following. • Drivers were more excited and less concentrated with tempo speeding up. Music can influence car following performance. However, it is not well resolved about its mediation effect on car following when the drivers' personalities are considered. We investigated how music style and tempo influence car following with different personalities. Twelve tracks were used in this study, four for each music tempo range, i.e., slow, medium, and fast tempo, and six for each music style, i.e., classical and pop one. The results showed introverts were more susceptible to music, and tend to listen to slow tempo music and classical one. In addition, pop music aroused the drivers more than classical and may induce closer headway distance. Furthermore, with the tempo speeding up, the drivers were more excited, less concentrated and performed less stablely. The medium music tempo was the most appropriate choice for keeping stable headway distance and taking actions to the changes of the leading vehicle. The present study shows personality can mediate the influence of music listening while driving, and music style and tempo can impact the mediation in a specific way. The study provides a guide on the music choice during driving and may bring benefits to the configuration of the music radio program and car music player. [ABSTRACT FROM AUTHOR]

Traffic light detection is a key module in the autonomous driving system to enhance the interactions between drivers and unmanned vehicles. In recent studies, deep neural networks are widely used for traffic light detection and resource/power consumption is a major concern for model deployment in vehicular edge devices. This paper proposes a novel light-weight deep CNN model that integrates the multi-backbone of state-of-the-art architectures for the self-driving traffic light detection. The MBBNet (Multi-BackBone Network) consists of three common convolutional backbones, i.e., the normal, residual and highway (DenseNet) convolutional modules. Simple ensemble of those backbones may incur high computational load. Therefore, channel compression is adopted to control the model parameters, while guaranteeing the accuracy for mobile and embedded hardware. Evaluation of a dataset collected from real road conditions demonstrate the robustness of our detection system, and it achieves higher accuracy (accuracy > 0.94 and A v e r a g e _ I O U > 74.05 %) for self-driving buses. In terms of resource consumption, the trained model size is 1.35 MB, and can process high-resolution images (1280 × 960) at 14 FPS (frames per second) on low-power edge devices. [ABSTRACT FROM AUTHOR]

Waves are fundamentally important for sediment re-suspension in estuary and coastal areas, especially for silty sediments, which can be easily suspended by waves, but the differential effects of swell and wind waves are still unclear. Integrated field observations were made from November 2012 to March 2013 including waves, currents, and suspending sediments on the offshore seabed of the Huanghe Delta to explore the mechanism of sediment re-suspension in silty coastal zones. During the five months of observation, there were more than 30 winter wind events that affected the study area and induced sediment re-suspension with varying suspended sediment concentration. The observed wave composition was separated into swell and wind waves using a bandpass filter. Results show that large swell (with significant height > 1.0 m) coming from the offshore direction (NE in our study area) dominated sediment resuspension in the coastal seabed due to the fact that this wind direction had the longest average fetch. Winds from the onshore direction usually had smaller swell due to their short fetches and caused limited sediment re-suspension. The residual currents caused by NE winds also transport larger sediment. An individual NE wind event could transport sediment 8–13.6 t/m2 and 5.1–8.2 t/m2 in directions parallel and perpendicular, respectively, to the isobaths, which is much higher than the sediment transportation during an individual NW wind event, which could transport 1.5–4 t/m2 and 0.6–5 t/m2 parallel and perpendicular, respectively, to the isobaths. Our research shows that large swell and the accompanying residual currents caused by NE winds (from offshore direction) are a vital driving force for sediment resuspension and transportation in the offshore zone. • The height of swell strongly affected sediment resuspension in coastal seabed. • Large swell coming from offshore (NE in our study area) dominated sediment resuspension events in offshore zone (water depth >10 m). • During and after swell, residual currents are the major dynamics that transport sediment from coastal areas. • Abstract. [ABSTRACT FROM AUTHOR]

It is an important problem to design resource-efficient neural architectures. One solution is adjusting the number of channels in each layer and the number of blocks in each network stage. This paper presents a novel framework named network adjustment which considers accuracy as a function of the computational resource (e.g., FLOPs or parameters), so that architecture design becomes an optimization problem and can be solved with the gradient-based optimization method. The gradient is defined as the resource utilization ratio (RUR) of each changeable module (layer or block) in a network and is accurate only in a small neighborhood of the current status. Therefore, we estimate it using Dropout, a probabilistic operation, and optimize the network architecture iteratively. The computational overhead of the entire process is comparable to that of re-training the final model from scratch. We investigate two versions of RUR where the resource usage is measured by FLOPs and latency. Experiments on standard image classification datasets and a few base networks including ResNet and EfficientNet demonstrate the effectiveness of our approach, which consistently outperforms the pruning-based counterparts. [ABSTRACT FROM AUTHOR]

Microsoft Kinect, a low-cost motion capture device, has huge potential in applications that require machine vision, such as human-robot interactions, home-based rehabilitation and clinical assessments. The Kinect sensor can track 25 key three-dimensional (3D) "skeleton" joints on the human body at 30 frames per second, and the skeleton data often have acceptable accuracy. However, the skeleton data obtained from the sensor sometimes exhibit a high level of jitter due to noise and estimation error. This jitter is worse when there is occlusion or a subject moves slightly out of the field of view of the sensor for a short period of time. Therefore, this paper proposed a novel approach to simultaneously handle the noise and error in the skeleton data derived from Kinect. Initially, we adopted classification processing to divide the skeleton data into noise data and erroneous data. Furthermore, we used a Kalman filter to smooth the noise data and correct erroneous data. We performed an occlusion experiment to prove the effectiveness of our algorithm. The proposed method outperforms existing techniques, such as the moving mean filter and traditional Kalman filter. The experimental results show an improvement of accuracy of at least 58.7%, 47.5% and 22.5% compared to the original Kinect data, moving mean filter and traditional Kalman filter, respectively. Our method provides a new perspective for Kinect data processing and a solid data foundation for subsequent research that utilizes Kinect. [ABSTRACT FROM AUTHOR]

An amendment to this paper has been published and can be accessed via a link at the top of the paper. [ABSTRACT FROM AUTHOR]

Helmet comfort has always been important for the evaluation of infantry equipment accessories and has for decades not been well addressed. To evaluate the stability and comfort of the helmet, this paper proposes a novel type of helmet comfort measuring device. Conventional pressure measuring devices can measure the pressure of flat surfaces well, but they cannot accurately measure the pressure of curved structures with large curvatures. In this paper, a strain-resistive flexible sensor with a slice structure was used to form a matrix network containing more than a 100 sensors that fit the curved surface of the head well. Raw data were collected by the lower computer, and the original resistance value of the pressure was converted from analog to digital by the A/D conversion circuit that converts an analog signal into a digital signal. Then, the data were output to the data analysis and image display module of the upper computer. The complex curved surface of the head poses a challenge for the appropriate layout design of a head pressure measuring device. This study is expected to allow this intuitive and efficient technology to fit other wearable products, such as goggles, glasses, earphones and neck braces. [ABSTRACT FROM AUTHOR]

We studied pore water pressure responses in silty seabed under random wave action through a series of experiments in a wide wave flume. Unlike previous experiments involving regular waves, we focus on random waves including wind-induced short waves and long waves so as to gain further insights into seabed responses and liquefaction risks posed by random waves. In particular, the study investigated how the secondary long waves that were induced by incident short wave groups affected the seabed responses. The test results revealed that these long waves could cause much larger seabed responses than the short waves (eight times larger in our flume tests). Although they had smaller wave heights than the short waves, the long waves were found to contribute much more significantly to the cumulative pore pressure than previously recognized. The likely reason is that the long waves are disproportionally effective in generating cumulative excess pore pressure, confirming qualitatively some of the earlier theoretical predictions. One of the implications from these research findings is that the existing design methods when applied to random waves could grossly underestimate liquefaction potential in silty sediment bed if either spectrum-based mean wave parameters or significant wave parameters were used. [ABSTRACT FROM AUTHOR]

In recent years, deep learning has been widely used in breast cancer diagnosis, and many high-performance models have emerged. However, most of the existing deep learning models are mainly based on static breast ultrasound (US) images. In actual diagnostic process, contrast-enhanced ultrasound (CEUS) is a commonly used technique by radiologists. Compared with static breast US images, CEUS videos can provide more detailed blood supply information of tumors, and therefore can help radiologists make a more accurate diagnosis. In this paper, we propose a novel diagnosis model based on CEUS videos. The backbone of the model is a 3D convolutional neural network. More specifically, we notice that radiologists generally follow two specific patterns when browsing CEUS videos. One pattern is that they focus on specific time slots, and the other is that they pay attention to the differences between the CEUS frames and the corresponding US images. To incorporate these two patterns into our deep learning model, we design a domain-knowledge-guided temporal attention module and a channel attention module. We validate our model on our Breast-CEUS dataset composed of 221 cases. The result shows that our model can achieve a sensitivity of 97.2% and an accuracy of 86.3%. In particular, the incorporation of domain knowledge leads to a 3.5% improvement in sensitivity and a 6.0% improvement in specificity. Finally, we also prove the validity of two domain knowledge modules in the 3D convolutional neural network (C3D) and the 3D ResNet (R3D). [ABSTRACT FROM AUTHOR]

This paper explores the practicability of bimanual asymmetric interaction and the factors affecting it when manipulating virtual three-dimensional objects through Leap, a novel hand motion capture device. This research has the potential to aid in the rehabilitation of people with upper limb function impairments. The participants were asked to enlarge a virtual three-dimensional box device to a predefined specified scale. Three influencing factors were addressed, i.e., task difficulty, task allocation and interactive mode. The results indicate that all factors have significant effects on movement time. However, analysis of variance tests shows that there are significant effects on the error rate and spatial patterns due to task difficulty and interactive mode, while no significant effects are found for task allocation. Additionally, task allocation is observed to have significant effects on the time of each hand from zooming phase onset to peak velocity. The action of the nondominant hand is coarser than that of the dominant hand. Interestingly, the velocities of both hands synchronized as the task difficulty level increased, even though the limbs moved at quite different speeds in the initial stage. This research provides insights into how one hand coordinates with the other in terms of the temporal aspects of movement kinematics and thus can help in designing rehabilitative devices that interact with the healthy hand. [ABSTRACT FROM AUTHOR]

A GSMem malware can attack a computer connected physically with no network. However, none of the existing techniques can detect GSMem attacks, up to now. To address this problem, this paper puts forward a new method based on Machine Learning (ML), including Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Boosted Tree (BT), Back-Propagation Neural Networks (BPNN) and Naive Bayes Classifier (NBC). At first, we use a large quantity of data in terms of frequencies and amplitudes of some electromagnetic waves to train our models. And then, we use the obtained models to predict that whether a GSMem attack occurs or not, according to a given frequency and amplitude. In a word, the GSMem intrusion detection problem is induced to a ML binary classification one, while the former problem is pending and the latter one has been solved. As a result, the former problem can be solved in principle in this way. The simulated experiments show that the new method is potential to detect a GSMem attack, with low False Positive Rates (FPR) and low False Negative Rates (FNR). [ABSTRACT FROM AUTHOR]

Underwater image enhancement (UIE) aims to generate clear images from low-quality underwater images. Due to the unavailability of clear reference images, researchers often synthesize them to construct paired datasets for training deep models. However, these synthesized images may sometimes lack quality, adversely affecting training outcomes. To address this issue, we propose to boost UIE with Diffusion Prior (UIEDP). It is a novel framework treating UIE as a posterior distribution sampling process of clear images conditioned on degraded underwater inputs. Specifically, UIEDP combines a pre-trained diffusion model capturing natural image priors with any existing UIE algorithm, leveraging the latter to guide conditional generation. The diffusion prior mitigates the drawbacks of inferior synthetic images, resulting in higher-quality image generation. Extensive experiments have demonstrated that our UIEDP yields significant improvements across various metrics, especially no-reference image quality assessment. And the generated enhanced images also exhibit a more natural appearance. • A novel diffusion-based framework UIEDP for boosting underwater image enhancement. • UIEDP makes full use of natural image priors introduced by the pre-trained diffusion model. • UIEDP can integrate any existing underwater image enhancement method and boost the performance. [ABSTRACT FROM AUTHOR]

As one of the most important tasks of natural language processing, textual emotion classification (TEC) aims to recognize and detect all emotions contained in texts. However, most existing methods are implemented using deep learning approaches, which may suffer from long training time and low convergence. Motivated by these challenges, in this paper, we provide a new solution for TEC by using cascading broad learning (CBL) and sentence embedding using a masked and permuted pre-trained language model (MPNet), named CBLMP. Texts are input into MPNet to generate sentence embedding containing emotional semantic information. CBL is adopted to improve the ability of feature extraction in texts and to enhance model performance for general broad learning, by cascading feature nodes and cascading enhancement nodes, respectively. The L-curve model is adopted to ensure the balance between under-regularization and over-regularization for regularization parameter optimization. Extensive experiments have been carried out on datasets of SMP2020-EWECT and SemEval-2019 Task 3, and the results show that CBLMP outperforms the baseline methods in TEC. [ABSTRACT FROM AUTHOR]

The low-rank matrix completion problem is fundamental in both machine learning and computer vision fields with many important applications, such as recommendation system, motion capture, face recognition, and image inpainting. In order to avoid solving the rank minimization problem which is NP-hard, several surrogate functions of the rank have been proposed in the literature. However, the matrix restored from the optimization problem based on the existing surrogate functions seriously deviates from the original one. In this paper, we first design a new non-convex Schatten capped $p$ p norm which generalizes several existing non-convex matrix norms and balances between the rank and the nuclear norm of the matrix. Then, a matrix completion method based on the Schatten capped $p$ p norm is proposed by exploiting the framework of the alternating direction method of multipliers. Meanwhile, the Schatten capped $p$ p norm regularized least squares subproblem is analyzed in detail and is solved explicitly. Finally, we evaluate the performance of the proposed matrix completion method based on extensive experiments in the field of image inpainting. All the experimental results demonstrate that the proposed method can indeed improve the accuracy of matrix completion compared with the existing methods. [ABSTRACT FROM AUTHOR]

Commercialization of self-driving applications requires precision and reliability of the perception system due to the highly dynamic and complex road environment. Early perception systems either rely on the camera or on LiDAR for moving obstacle detection. With the development of vehicular sensors and deep learning technologies, the multi-view and sensor fusion based convolutional neural network (CNN) model for detection tasks has become a popular research area. In this paper, we present a novel multi-sensor fusion-based CNN model–SaccadeFork–that integrates the image and upsampled LiDAR point clouds as the input. SaccadeFork includes two modules: (1) a lightweight backbone that consists of hourglass convolution feature extraction module and a parallel dilation convolution module for adaptation of the system to different target sizes; (2) an anchor-based detection head. The model also considers deployment of resource-limited edge devices in the vehicle. Two refinement strategies, i.e., Mixup and Swish activation function are also adopted to improve the model. Comparison with a series of latest models on public dataset of KITTI shows that SaccadeFork can achieve the optimal detection accuracy on vehicles and pedestrians under different scenarios. The final model is also deployed and tested on a local dataset collected based on edge devices and low-cost sensor solutions, and the results show that the model can achieve real-time efficiency and high detection accuracy. [Display omitted] • Fusion of LiDAR point cloud and image can improve detection performance of CNN. • Two interpolation algorithms are used to enhance point cloud feature. • A light weight CNN model consists of hourglass and parallel dilated convolution. • SaccadeFork is tested on both the public dataset KITTI and local platform. • Low-cost multi-sensor fusion can achieve similar performance of a Velodyne-64E. [ABSTRACT FROM AUTHOR]

Numerous estuaries worldwide have been modified in the past decades by human interventions. The shoreline and bathymetry in the Pearl River Estuary (PRE) have changed greatly over the past 40 years due to the influence of land reclamation and waterway dredging, which have resulted in the corresponding adjustment of its hydrodynamics. Utilizing the ROMS numerical model, this paper studies the hydrodynamic responses to changes in shoreline and bathymetry in the PRE from 1971 to 2012. The results show that, on one hand, during neap tide, the change in the shoreline makes the residual current in the West Channel (WC) increase by 0.10 (0.05) m/s at maximum in the surface (near bottom) layer. Therefore, the exchange flow increases by 9.5% and the longitudinal circulation strengthens. The surface isohalines move southward up to 18 km, but the bottom isohalines move northward ∼2 km in the WC and East Channel (EC), which is different from the previous conclusion that the seaward extension of coastlines inhibits saltwater intrusion. The decrease in salinity in the upper layer reduces the upper seaward salt transport, resulting in a larger net landward salt transport from 25.21 × 103 kg/s to 35.44 × 103 kg/s. During spring tide, the changes are relatively weaker, but the direction of salt transport changes to seaward and the net seaward transport also increases. Moreover, the change in shoreline reduces the water area and volume in the PRE by 21.3% and 15.6% respectively, which causes a reduction of 11.3% (17.1%) in tidal prism during spring (neap) tide. The wave celerity is enhanced (>23%) and the amplification of semidiurnal tide is strengthened (>20%) in the WC and West Shoal (WS). The reduction in tidal prism together with the strengthened reflection of tidal waves with a larger phase lag between elevation and velocity of M 2 tidal component cause a significant decrease of 19.0% in tidal energy flux entering the PRE. However, the tidal range increases by ∼0.30 m (mainly due to the increase in M 2 tidal amplitude) in the EC , which is induced by lower tidal energy dissipation there. On the other hand, during neap tide, the change in bathymetry greatly enhances the bottom landward residual current whose peak value increases by 80% in the WC , thereby increasing approximately 14 km of the intrusion distance of saline water, enhancing the exchange flow by 27.5% and strengthening the longitudinal circulation. The net landward salt transport increases to 38.39 × 103 kg/s. The results during spring tide are similar to those during neap tide but with smaller changes, and the net seaward salt transport decreases. Moreover, the wave celerity is slightly reduced (<8%) and the amplification of semidiurnal tide is also decreased (<10%) in the WC and WS. Meanwhile, the change in bathymetry only decreases the water volume by 4.7%, leading to a relatively smaller effect on the tidal prism, and it strengthens (weakens) the reflection of tidal waves in the West Shoal and Middle Shoal (WC and EC), resulting in a slight reduction in tidal energy flux entering the bay. In the WC , the tidal range is basically unchanged since the increased tidal energy flux is offset by the increased dissipation. The quantitative results obtained in this study may provide some references for the development and protection of the PRE and other estuaries that are subject to strong human interventions. • Seaward extension of coastlines increases tidal range, wave celerity and saltwater intrusion. • Change in bathymetry increases ∼14 km of the intrusion distance of saltwater. • Change in bathymetry enhances exchange flow by 27.5%, reduces tidal energy flux by 2.4%. [ABSTRACT FROM AUTHOR]

With the popularity of social networks, sentiment analysis has become one of the hottest topics in natural language processing (NLP). As the development of research on the fine-grained sentiment analysis, more and more researchers pay attention to aspect-level sentiment analysis. It aims to identify the same or different sentiment polarity in different aspects of the context. In this paper, a context and aspect memory network (CAMN) method is proposed to solve the problem of aspect level sentiment analysis. In this method, deep memory network, bi-directional long short-term memory network and multi-attention mechanism are introduced to better capture the sentiment features in short texts. It includes two strategies: one is to use the self-attention mechanism (i.e., CAMN-SA) to calculate the context relevance; the other is to use the encoder-decoder attention mechanism (i.e., CAMN-ED) to calculate the context and aspect relevance. In order to verify the function of each component in the proposed method, and to test the effect of different hops on the memory network, we conduct many experiments on three real-world datasets to compare the baseline models with our proposed method. Experimental results show that our proposed method can achieve better performance than the baseline models. [ABSTRACT FROM AUTHOR]

Numerous research efforts have been conducted to simulate the crowd movements, while relatively few of them are specifically focused on multihazard situations. In this paper, we propose a novel crowd simulation method by modeling the generation and contagion of panic emotion under multihazard circumstances. In order to depict the effect from hazards and other agents to crowd movement, we first classify hazards into different types (transient and persistent, concurrent and nonconcurrent, and static and dynamic) based on their inherent characteristics. Second, we introduce the concept of perilous field for each hazard and further transform the critical level of the field to its invoked-panic emotion. After that, we propose an emotional contagion model to simulate the evolving process of panic emotion caused by multiple hazards. Finally, we introduce an emotional reciprocal velocity obstacles (RVOs) model to simulate the crowd behaviors by augmenting the traditional RVO model with emotional contagion, which for the first time combines the emotional impact and local avoidance together. Our experimental results demonstrate that the overall approach is robust, can better generate realistic crowds and the panic emotion dynamics in a crowd. Furthermore, it is recommended that our method can be applied to various complex multihazard environments. [ABSTRACT FROM AUTHOR]

• Adopting neural networks to predict channel water level can improve accuracies. • Features of multiple channels and time-slices are helpful for water level prediction. • The prediction model based on recurrent neural network is more effective. • The accuracy of recurrent neural network based model reaches up to 97.05%. Water level prediction is crucial to water diversion through cascaded channels, and the prediction accuracies are still unsatisfying due to the difficulties and challenges caused by complex interactions and relations among cascaded channels. We adopt two kinds of neural networks to build our water level prediction models for cascaded channels 2/4/6 h ahead with high prediction accuracy. First, the raw hydrological data of cascaded channels are augmented using spatial and temporal windows, which produces data sets with high-dimensional features. Then, Multilayer Perceptron (MLP) and Recurrent Neural Network (RNN) are adopted to build the water level prediction model with the help of the augmented data containing the implicit correlation among multiple channels in spatial dimension and multiple data records in temporal dimension. China's South-to-North Water Diversion Project is taken as the case study. Experimental results show that our models outperform Support Vector Machine (SVM) by 34.78%, 44.53%, 1.32% and 9.198% in terms of Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Pearson Correlation Coefficient (PCC) and Nash' Sutcliffe Efficiency(NSE), respectively. The accuracies of our models with prediction deviations less than 1 cm, 2 cm, and 3 cm can reach as high as 81.36%, 94.09%, and 97.05%, respectively. [ABSTRACT FROM AUTHOR]

• Linear phase speed and dispersive coefficient of ISWs modulated by mean Sulu Sea circulation decrease, while nonlinear coefficient increases. • Seasonal cycles of two leading modes of ISW dynamic parameters reflect the modulation of Sulu Sea basin-scale circulation and meander flow, respectively. • Peak-to-peak fluctuations of linear phase speed, nonlinear and dispersive coefficients reach 15%, 30% and 20%, respectively, of those due to topography. In-situ observations of the Sulu Sea internal solitary waves (ISWs) are too limited to reveal their variability. In this study, based on verified model reanalysis data, we study the variability of three basic dynamic parameters of ISWs, i.e., linear phase speed, nonlinear and dispersive coefficients, affected by the Sulu Sea circulations. Overall, the linear speed and dispersive coefficient of ISWs modulated by the annual-mean Sulu Sea circulation decrease, while the nonlinear coefficient increases, and these variations are largely attributed to the variable stratification rather than current. Particularly, at the deep basin and near the western boundary of the Sulu Sea, the modulated climatological annual-mean linear phase speed reduces at least ∼10% of that resulted from only topography, while at the deep basin the modulated dispersive (nonlinear) coefficient has a significant decrease (increase) up to approximately 25–30% (20–30%) relative to that resulted from only topography. Moreover, the first EOF mode of the three ISW dynamic parameters, dominant in winter and summer, reflects the modulation of the Sulu Sea basin-scale circulation, while the second EOF mode, dominant in spring and autumn, illustrates the modulation associated with the meander flow along the western boundary. Specifically, it is found that the seasonal cycles of linear phase speed (nonlinear and dispersive coefficients) are largely attributed to the variable current (stratification) induced by the seasonal circulations. The circulation-induced peak-to-peak fluctuations of linear speed, nonlinear and dispersive coefficients reach up to ∼15%, ∼30% and ∼20%, respectively, of those resulted from only topography. [ABSTRACT FROM AUTHOR]

Existing domain adaptation methods for classifying textual emotions have the propensity to focus on single-source domain exploration rather than multi-source domain adaptation. The efficacy of emotion classification is hampered by the restricted information and volume from a single source domain. Thus, to improve the performance of domain adaptation, we present a novel multi-source domain adaptation approach for emotion classification, by combining broad learning and deep learning in this article. Specifically, we first design a model to extract domain-invariant features from each source domain to the same target domain by using BERT and Bi-LSTM, which can better capture contextual features. Then we adopt broad learning to train multiple classifiers based on the domain-invariant features, which can more effectively conduct multi-label classification tasks. In addition, we design a co-training model to boost these classifiers. Finally, we carry out several experiments on four datasets by comparison with the baseline methods. The experimental results show that our proposed approach can significantly outperform the baseline methods for textual emotion classification. [ABSTRACT FROM AUTHOR]

Wave-induced pore pressure in the seabed may cause seabed liquefaction and lead to geohazards such as submarine landslides. Under natural conditions, waves generally consist of different components with various frequencies, and the seabed response to each wave component can vary considerably. However, little is known concerning these differences. In this study, a series of flume experiments were conducted to investigate the pore pressure responses in a silty seabed to various components of random waves. According to the experimental results, the short wave (SW) component had high energies, but the SW-generated high-frequency hydrodynamic pressure (HF p0) decayed rapidly in the seabed. The long wave (LW) component had low energies, but the LW-generated low-frequency hydrodynamic pressure (LF p0) was effectively propagated in the seabed. The transmittance coefficient of the low-frequency pore pressures (LF p) was two times that of the high-frequency pore pressures (HF p). The energy ratio of LF p /HF p was enhanced with increasing soil depth, and the energy of LF p was greater than that of HF p at soil depths ≥15 cm, indicating that the contribution of LWs to the cumulative pore pressure is enhanced with an increase in soil depth. An important implication of our findings is that the use of statistical wave parameters (e.g., significant wave height and average wave period) to determine pore pressure responses of the seabed can underestimate liquefaction risk, and in turn, lead to unsafe engineering designs. • Long wave component generates significantly stronger pore pressure. • High-frequency pore pressures propagated poorly in the seabed. • Low-frequency pore pressures readily propagated within the seabed. • Long wave component had significantly larger contribution to deep soil pore pressure. [ABSTRACT FROM AUTHOR]

Background: Assessing working posture risks is important for occupational safety and health. However, low-cost assessment techniques for human motion injuries in the logistics delivery industry have rarely been reported., Objective: To propose a novel approach for posture risk assessment using low-cost motion capture with artificial intelligence., Methods: A Kinect was adopted to obtain red-green-blue (RGB) and depth images of the subject with 24 postures, and the human joints were extracted using artificial intelligence. The images were registered to obtain the actual three-dimensional (3D) human joint angle., Results: The root mean square error (RMSE) significantly decreased. Finally, two common methods for evaluating human working posture injuries-the Rapid Upper Limb Assessment and Ovako Working Posture Analysis System-were investigated., Conclusions: The outputs of the proposed method are consistent with those of the commercial ergonomic evaluation software.

Background: The physical factors associated with musculoskeletal pain in nursing personnel have been largely investigated, although the role of sleep and psychological factors resulting in musculoskeletal pain has not been addressed thoroughly., Objective: This study aimed to explore the prevalence of musculoskeletal pain and investigate how sleep and psychological factors influence musculoskeletal pain in a nursing group., Methods: Nordic standard questionnaires were distributed to 230 female nurses. Chi-square tests were performed to assess the associations between sleep problems, psychological problems, and musculoskeletal pain symptoms. Binary logistic regression analysis was also conducted to identify the primary factors influencing the prevalence of musculoskeletal pain., Results: The highest prevalence of pain was observed in the lower back, neck, and shoulders, whereas the lowest prevalence of pain was observed in the ankles, feet, elbows, and hips/buttocks. Chi-square analysis and binary logistic regression showed that sleep duration, sleep onset time, and sleep quality all significantly contributed to the development of neck and upper back pain. With regard to the psychological factors, only occupational pride and stress had a significant effect on pain; in contrast, family support did not show any significant influence., Conclusion: Compared with other body regions, musculoskeletal pain in the lower back, neck, and shoulders requires more attention and preventive interventions. Special efforts should be made to shift the workday system of the nursing group because of the strong correlation between sleep problems and pain. Incentives other than penalty mechanisms should be considered seriously in nursing to boost occupational pride and relieve job stress.

Experienced radiologists can accurately diagnose relevant diseases by observing the cardiopulmonary region in chest X-ray images. Advances in deep learning techniques enable the prediction of lesions in chest X-ray images. However, deep learning-based algorithms usually require a large amount of training data, and it lacks an effective method for data generation and augmentation. In this paper, we propose a Lung Segmentation Reconstruction (LSR) module. A healthy chest X-ray image is generated based on the abnormal image as a reference. With the generated healthy reference, we propose a novel way of data augmentation for chest X-ray images. The whole images, lung regions and abnormal regions are stacked together and fed into a classification model to make a credible diagnosis. Extensive experiments have been conducted on the public dataset CheXpert. Experimental results show that our proposed abnormality enhancement can help the baseline models achieve better performance on consolidation and pleural effusion. These results highlight the potential value of the large number of healthy chest X-ray images in the dataset and the combination of different regions of chest X-ray images for prediction.

Accurate breast lesion segmentation in ultrasound images helps radiologists to make exact diagnoses and treatments, which is important to increase the survival rate of breast cancer patients. Recently, deep learning-based methods have demonstrated remarkable results in breast lesion segmentation. However, the blurry breast lesion boundaries and noise artifacts in ultrasound images still limit the performance of the deep learning-based methods. In this paper, we propose a novel segmentation network equipped with a focal self-attention block for improving the performance of breast lesion segmentation. The focal self-attention block can incorporate fine-grained local and coarse-grained global information. The fine-grained local information is useful to enhance features of breast lesion boundaries, while the coarse-grained global information effectively reduces noise interference. To verify the performance of our network, we implement breast lesion segmentation on our collected dataset of 9836 ultrasound images. The results demonstrate that the focal self-attention block enhances features of breast lesion boundaries and improves the accuracy of breast lesion segmentation.

Objectives: Machine learning is increasingly being used in the medical field. Based on machine learning models, the present study aims to improve the prediction performance of craniodentofacial morphological harmony judgment after orthodontic treatment and to determine the most significant factors. Methods: A dataset of 180 subjects was randomly selected from a large sample of 3,706 finished orthodontic cases from six top orthodontic treatment centers around China. Thirteen algorithms were used to predict the value of the cephalometric morphological harmony score of each subject and to search for the optimal model. Based on the feature importance ranking and by removing features, the regression models of machine learning (including the Adaboost, ExtraTree, XGBoost, and linear regression models) were used to predict and compare the score of harmony for each subject from the dataset with cross validations. By analyzing the prediction values, the most optimal model and the most significant cephalometric characteristics were determined. Results: When nine features were included, the performance of the XGBoost regression model was MAE = 0.267, RMSE = 0.341, and Pearson correlation coefficient = 0.683, which indicated that the XGBoost regression model exhibited the best fitting and predicting performance for craniodentofacial morphological harmony judgment. Nine cephalometric features including L1/NB (inclination of the lower central incisors), ANB (sagittal position between the maxilla and mandible), LL-EP (distance from the point of the prominence of the lower lip to the aesthetic plane), SN/OP (inclination of the occlusal plane), SNB (sagittal position of the mandible in relation to the cranial base), U1/SN (inclination of the upper incisors to the cranial base), L1-NB (protrusion of the lower central incisors), Ns-Prn-Pos (nasal protrusion), and U1/L1 (relationship between the protrusions of the upper and lower central incisors) were revealed to significantly influence the judgment. Conclusion: The application of the XGBoost regression model enhanced the predictive ability regarding the craniodentofacial morphological harmony evaluation by experts after orthodontic treatment. Teeth position, teeth alignment, jaw position, and soft tissue morphology would be the most significant factors influencing the judgment. The methodology also provided guidance for the application of machine learning models to resolve medical problems characterized by limited sample size., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2022 Wang, Zhao, Song, Niu and Xu.)

Objective: The driver's instantaneous situation awareness in the process of take-over of vehicle control in automated driving has not yet been thoroughly investigated. The proposed research can provide a better understanding of the driver's perceived characteristics and identify the most urgent information requirements of the on-site scenario when the driver's eye sight returns from other distractors to the driving scene., Methods: We conducted an experiment in simulated automated driving to study the participants' ability of instantaneous hazard perception and judgment. The scene pictures, which were displayed in millisecond time, were used to imitate the situations that drivers would see when the distracted drivers returned their gaze from the distractive sources to the road in the simulated automated driving., Results: The results show that the driving state, scene representation time and hazard levels affect the instantaneous situation awareness of drivers. In addition, the scene perception accuracy of the group who played games during automated driving is much lower than that of the group that chatted with the copilot. The longer picture-showing duration decreases the accuracy of hazard identification, whereas the shorter picture-showing duration increases the accuracy of hazard perception and the hazard rating score., Conclusions: In conclusion, distraction reduces the accuracy of the instantaneous scene perception of drivers, and drivers behave more cautiously in decision making when the driving situations are more hazardous. This study provides a good theoretical basis for the design of hazard warning information for automated driving.

Eye-tracking has been a hot topic in human-computer interaction (HCI). Nevertheless, previous studies usually adopted eye-tracking as information output rather than input. The eye-control technique can achieve convenient and rapid real-time operation through the movement of the eyes and reduce unnecessary manual operations. Because the layout determines the location orientation, organizational complexity, cognitive consistency, and predictive ability of the information display, the interface layout design affects the user's perception of information intensity, complexity, and logic. Moreover, the method of target clicking by eye-control techniques, which include blink and dwell, also depends on the application and user's ability. The purpose of this study is to investigate the influence of target layout and target picking method on picking time and dragging performance based on eye-control technique. The results indicate that the target picking method, i.e., blink or dwell, had significant effects on the dragging time and dragging numbers. However, there was no significant effect of target layout on picking time and dragging performance (dragging time and numbers), which may be related to the setting of the experimental conditions (e.g., lighting level and screen resolution). Moreover, the target picking method and the target layout had no significant interaction effect on picking time and dragging performance. The findings are anticipated to provide helpful implications for future eye control technique design., (Copyright © 2020 Wang, Wang, Zhou, Liu, Shi, Zhao, Zhang and Niu.)

BACKGROUND: Manual rendezvous and docking (RVD) is challenging for the astronauts, and automation is used to aid this operation. However, the automation mode in the final approaching stance of RVD is quite different. This paper is aimed at investigating the effect of automation on performance, workload and situation awareness (SA) among novice and expert operators in RVD. METHODS: A two-factor mixed experimental design was adopted. There were 15 novices and 12 experts who participated in the experiment. All subjects were required to finish six tasks of two automation levels: manual RVD and automation-aided RVD. The Performance was assessed by docking result and control process. Workload and SA were measured by NASA Task Load Index and Situation Awareness Rating Techniques (SART). Repeat measures ANOVA and the simple effect test were used to analyze the effect of automation, skill level, and the interaction between them on performance, workload, and SA of operators. RESULTS: Novices exhibited performances inferior to experts, but the skills gap was attenuated as automation was introduced. Moreover, automation can enhance performance, reduce workload, and enhance SA for novices, but potentially deteriorate task performance and SA for the experienced. Mediation analysis results indicated automation was a significant predictor of workload and SA, b = -0.576 and b = 0.503, and workload and SA were significant predictors of docking result, b = -0.590 and b = 0.348. CONCLUSION: Automation can be detrimental to various elements of the functioning of highly experienced operators. Moreover, automation affects docking result by affecting workload and SA. Du X, Niu J, Zhang Y, Wang M, Wang D, Wu B, Cai J, Huang W. Performance, workload, and situation awareness in manual and automation-aided rendezvous and docking. Aerosp Med Hum Perform. 2019; 90(5):447-455.

Background: Stumbles are common accidents that can result in falls and serious injuries, particularly in the workplace where back and forth movements are involved and in offices where high heels are imperative. Currently, the characteristics of plantar pressure during a stumble and the differences between stumbling and a normal gait remain unclear., Objective: This paper is aimed at providing insights into the feasibility of the data mining technique for interventions in stumble-related occupational safety issues., Methods: The characteristics of plantar pressure distribution during stumbling and normal gait were analyzed by using the power spectrum density (PSD) and the Support Vector Machine (SVM). The PSD, a novel pattern recognition feature, was used to mathematically describe the image signal. The SVM, a powerful data mining technique, was used as the classifier to recognize a stumble. Dynamic plantar pressures were measured from twelve healthy participants as they walked., Results: The plantar pressures of the stumbling gaits had significantly different patterns compared to the normal ones, from either a qualitative or quantitative perspective. The mean recognition accuracy of the proposed method reached 96.7%., Conclusions: This study helps better understand stumbles and provides a theoretical basis for stumble-related occupational injuries. In addition, the stumble is the precursor of a fall and the research on stumble recognition would be of value to predict and provide warnings of falls and to design anti-fall devices for potential victims.

Objective: This study explores the influence of mobile phone secondary tasks on driving from the perspective of visual, auditory, cognitive, and psychomotor (VACP) multiple resource theory, and it is anticipated to benefit the human-centered design of mobile phone use while driving., Methods: The present study investigated 6 typical phone use scenarios while driving and analyzed the effects of phone use distractions on driving performance. Thirty-six participants were recruited to participate in this experiment. We abandoned traditional secondary tasks such as conversations or dialing, in which cognitive resources can become interference. Instead, we adopted an arrow secondary task and an n-back delayed digit recall task., Results: The results show that all mobile phone use scenarios have a significant influence on driving performance, especially on lateral vehicle control. The visual plus psychomotor resource occupation scenario demonstrated the greatest deterioration of driving performance, and there was a significant deterioration of driving speed and steering wheel angle once the psychomotor resource was occupied., Conclusions: Phone use distraction leads to visual, cognitive, and/or motor resource functional limitations and thus causes lane violations and traffic accidents.

In complex networks, cluster structure, identified by the heterogeneity of nodes, has become a common and important topological property. Network clustering methods are thus significant for the study of complex networks. Currently, many typical clustering algorithms have some weakness like inaccuracy and slow convergence. In this paper, we propose a clustering algorithm by calculating the core influence of nodes. The clustering process is a simulation of the process of cluster formation in sociology. The algorithm detects the nodes with core influence through their betweenness centrality, and builds the cluster's core structure by discriminant functions. Next, the algorithm gets the final cluster structure after clustering the rest of the nodes in the network by optimizing method. Experiments on different datasets show that the clustering accuracy of this algorithm is superior to the classical clustering algorithm (Fast-Newman algorithm). It clusters faster and plays a positive role in revealing the real cluster structure of complex networks precisely.