A Self-Supervised Evaluation Approach of Insulation Condition for Vehicle Cable Terminals Using Hypergraph Neural Network With Dynamic Features

Visual information consists of static and dynamic properties. How is their representation organized in the visual system? Static information has been associated with ventral temporal regions while dynamic information with lateral and dorsal regions. Investigating the representation of static and dynamic information is complicated by the correlation between static and dynamic information within continuous visual input. Here, we used two-stream deep convolutional neural networks (DCNNs) to separate static and dynamic features in quasi-naturalistic videos and to investigate their neural representations. The first DCNN stream was trained to represent static features by recognizing action labels using individual video frames, and the second DCNN stream was trained to encode dynamic features by recognizing actions from optic flow information that describes changes across different frames. To investigate the representation of these different types of features in the visual system, we used representational similarity analysis to compare the neural network models to the neural responses in different visual pathways of 14 human participants (six females). First, we found that both static and dynamic features are encoded across all visual pathways. Second, we found that distinct visual pathways represent overlapping as well as unique static and dynamic visual information. Finally, multivariate analysis revealed that ventral and dorsal visual pathways share a similar posterior-to-anterior gradient in the representation of static and dynamic visual features.

Face liveness detection is a significant research topic in face-based online authentication. The current face liveness detection approaches utilize either static or dynamic features, but not both. In fact, the dynamic and static features have different advantages in face liveness detection. In this paper, we propose a scheme combining dynamic and static features to capture merits of them for face liveness detection. First, the dynamic maps are captured from the inter-frame motion in the video, which investigates motion information of the face in the video. Then, with a Convolutional Neural Network (CNN), the dynamic and static features are extracted from the dynamic maps and the frame images, respectively. Next, in CNN, the fully connected layers containing the dynamic and static features are concatenated to form a fused feature. Finally, the fused features are used to train a binary Support Vector Machine (SVM) classifier, which classifies the frames into two categories, i.e. frame with real or fake face. Experimental results and the corresponding analysis demonstrate that the proposed scheme is capable of discovering face liveness by fusing dynamic and static features and it outperforms the current state-of-the-art face liveness detection approaches.

Face liveness detection is an interesting research topic in face-based online authentication. The current face liveness detection algorithms utilize either static or dynamic features, but not both. In fact, the dynamic and static features have different advantages in face liveness detection. In this paper, we discuss a scheme to combine dynamic and static features that combines the strength of each. First, the dynamic maps are obtained from the inter frame motion in the video. Then, using a Convolutional Neural Network (CNN), the dynamic and static features are extracted from the dynamic maps and the images, respectively. Next, the fully connected layers from the CNN that include the dynamic and static features are connected to form the fused features. Finally, the fused features are used to train a two-value Support Vector Machine (SVM) classifier, which classify the images into two groups, images with real faces and images with fake faces. We conduct experiments to assess our algorithm that includes classifying images from two public databases. Experimental results demonstrate that our algorithm outperforms current state-of-the-art face liveness detection algorithms.

ATDA: Attentional temporal dynamic activation for speech emotion recognition

Fire is a fierce disaster, and smoke is the early signal of fire. Since such features as chrominance, texture, and shape of smoke are very special, a lot of methods based on these features have been developed. But these static characteristics vary widely, so there are some exceptions leading to low detection accuracy. On the other side, the motion of smoke is much more discriminating than the aforementioned features, so a time-domain neural network is proposed to extract its dynamic characteristics. This smoke recognition network has these advantages:(1) extract the spatiotemporal with the 3D filters which work on dynamic and static characteristics synchronously; (2) high accuracy, 87.31% samples being classified rightly, which is the state of the art even in a chaotic environments, and the fuzzy objects for other methods, such as haze, fog, and climbing cars, are distinguished distinctly; (3) high sensitiveness, smoke being detected averagely at the 23rd frame, which is also the state of the art, which is meaningful to alarm early fire as soon as possible; and (4) it is not been based on any hypothesis, which guarantee the method compatible. Finally, a new metric, the difference between the first frame in which smoke is detected and the first frame in which smoke happens, is proposed to compare the algorithms sensitivity in videos. The experiments confirm that the dynamic characteristics are more discriminating than the aforementioned static characteristics, and smoke recognition network is a good tool to extract compound feature.

In general, mechanical equipment such as cars, airplanes, and machine tools all operate with constant frequency characteristics. These constant working characteristics should be controlled if the dynamic performance of the equipment demands improvement or the dynamic characteristics is intended to change with different working conditions. Active control is a stable and beneficial method for this, but current active control methods mainly focus on vibration control for reducing the vibration amplitudes in the time domain or frequency domain. In this paper, a new method of dynamic frequency characteristics active control (DFCAC) is presented for a flat plate, which can not only accomplish vibration control but also arbitrarily change the dynamic characteristics of the equipment. The proposed DFCAC algorithm is based on a neural network including two parts of the identification implement and the controller. The effectiveness of the DFCAC method is verified by several simulation and experiments, which provide desirable results.

Predicting human future motion holds significant importance in the domains of autonomous driving and public safety. Kinematic features, including joint coordinates and velocity, are commonly employed in skeleton-based human motion prediction. Nevertheless, most existing approaches neglect the critical role of dynamic information and tend to degrade as the prediction length increases. To address the related constraints due to single-scale and fixed-joint topological relationships, this study proposes a novel method that incorporates joint torques estimated via Lagrangian equations as dynamic features of the human body. Specifically, the human skeleton is modeled as a multi-rigid body system, with generalized joint torques calculated based on the Lagrangian formula. Furthermore, to extract both kinematic and dynamic joint information effectively for predicting long-term human motion, we propose a Multiscale Mixed-Graph Neural Network (MS-MGNN). MS-MGNN can extract kinematic and dynamic joint features across three distinct scales: joints, limbs, and body parts. The extraction of joint features at each scale is facilitated by a single-scale mixed-graph convolution module. And to effectively integrate the extracted kinematic and dynamic features, a KD-fused Graph-GRU (Kinematic and Dynamics Fused Graph Gate Recurrent Unit) predictor is designed to fuse them. Finally, the proposed method exhibits superior motion prediction capabilities across multiple motions. In motion prediction experiments on the Human3.6 dataset, it outperforms existing approaches by decreasing the average prediction error by 9.1%, 12.2%, and 10.9% at 160 ms, 320 ms, and 400 ms for short-term prediction and 7.1% at 560 ms for long-term prediction.

Nanopore technology holds great potential for single-molecule identification. However, extracting meaningful features from ionic current signals and understanding the molecular mechanisms underlying the specific features remain unresolved. In this study, we uncovered a distinctive ionic current pattern in a K238Q aerolysin nanopore, characterized by transient spikes superimposed on two stable transition states. By employing a neural network model, we demonstrated that these previously overlooked dynamic spike features exhibit superior discriminative power, improving the accuracy from 44% to 93%. We identified that the stable transition states result from simultaneous interactions of ssDNA with the two sensitive sites of the nanopore. The proposed stochastic collision model offers a mechanistic framework for interpreting the generation of the dynamic spike features. This model indicates that the continuous transitions facilitate iterative, comprehensive snapshots of molecular interactions by nanopores. Our findings introduce a new approach for optimizing nanopore technology to capture complex dynamic features and substantially improve the accuracy of single-molecule identification.

SDTF-Net: Static and dynamic time–frequency network for Speech Emotion Recognition

Research on a monitoring model of revolute pair clearance based on dynamic features and thermal imaging fusion

A spoken word recognition method using dynamic features of speech and neural networks is presented. Dynamic features of speech are obtained from a two-dimensional mel-cepstrum (TDMC). The TDMC is defined as the two-dimensional Fourier transform of mel-frequency scaled log spectra in the frequency and time domains. It has averaged spectral features, dynamic spectral features, and averaged and dynamic features of power of the two-dimensional mel-log spectra in the analyzed interval. The neural network in this study is a three-layered feedforward neural network and learns automatically using a back-propagation algorithm. Dynamic spectral features, and averaged and dynamic features of power are used as the input of a neural network. The experimental results of speaker-dependent word recognition experiments for 100 Japanese city names uttered by nine speakers show that dynamic spectral features smoothed with respect to time are effective, and a recognition accuracy of 99.1% was obtained. >

This study presents an effective means of applying neural networks to achieve robust design with dynamic characteristic considerations. Two neural networks are constructed to train the data set in the Taguchi’s orthogonal array (OA): one to search for the optimal condition, and the other to forecast the system’s response value. A measuring system employed in semiconductor manufacturing demonstrates the proposed approach’s effectiveness. According to those results, the proposed approach outperforms the conventional Taguchi method. By using the proposed approach, the adjustment factors are not a prerequisite for the dynamic characteristic problem. Moreover, the proposed approach enhances the generalization capability.

Emotion perception is person-dependent and variable. Dimensional characterizations of emotion can capture this variability by describing emotion in terms of its properties (e.g., valence, positive vs. negative, and activation, calm vs. excited). However, in many emotion recognition systems, this variability is often considered "noise" and is attenuated by averaging across raters. Yet, inter-rater variability provides information about the subtlety or clarity of an emotional expression and can be used to describe complex emotions. In this paper, we investigate methods that can effectively capture the variability across evaluators by predicting emotion perception as a discrete probability distribution in the valence-activation space. We propose: (1) a label processing method that can generate two-dimensional discrete probability distributions of emotion from a limited number of ordinal labels; (2) a new approach that predicts the generated probabilistic distributions using dynamic audio-visual features and Convolutional Neural Networks (CNNs). Our experimental results on the MSP-IMPROV corpus suggest that the proposed approach is more effective than the conventional Support Vector Regressions (SVRs) approach with utterance-level statistical features, and that feature-level fusion of the audio and video modalities outperforms decision-level fusion. The proposed CNN model predominantly improves the prediction accuracy for the valence dimension and brings a consistent performance improvement over data recorded from natural interactions. The results demonstrate the effectiveness of generating emotion distributions from limited number of labels and predicting the distribution using dynamic features and neural networks.

Evaluation approach to dynamic characteristic of turbine flowmeters considering calibration system response

Magnetic rheological (MR) damper, as today's new semi-active control device, is widely used in vibration control engineering. However, in most control methods the controller's dynamic characteristics need to be known in advance. Because of highly nonlinear characteristics of MR damper, it is very difficult to establish its mathematical model to describe the reverse dynamic characteristics, which is essential in achieving the overall control strategy. In this paper, based on the identification role of neural network in complex nonlinear systems, according to performance tests of MR damper, the dynamic and inverse dynamic characteristic neural network model of MR damper is established, and the analysis and comparison of the neural network model conclusions and experimental conclusions are given, the results show that the neural network model of MR damper dynamic characteristics is reliable and effective.