Search

Presently available speech emotion recognition (SER) methods generally rely on a single SER model. Getting a higher accuracy of SER involves feature extraction method and model design scheme in the speech. However, the generalization performance of models is typically poor because the emotional features of different speakers can vary substantially. The present work addresses this issue by applying a two-level discriminative model to the SER task. The first level places an individual speaker within a specific speaker group according to the speaker’s characteristics. The second level constructs a personalized SER model for each group of speakers using the wave field dynamics model and a dual-channel general SER model. Two-level discriminative model are fused for implementing an ensemble learning scheme to achieve effective SER classification. The proposed method is demonstrated to provide higher SER accuracy in experiments based on interactive emotional dynamic motion capture (IEMOCAP) corpus and a custom-built SER corpus. In IEMOCAP corpus, the proposed model improves the recognition accuracy by 7%. In custom-built SER corpus, both masked and unmasked speakers is employed to demonstrate that the proposed method maintains higher SER accuracy.

As one of the most challenging and promising topics in speech field, emotion speech synthesis is a hot topic in current research. At present, the emotion expression ability, synthesis speed and robustness of synthetic speech need to be improved. Cycle-consistent Adversarial Networks (CycleGAN) provides a two-way breakthrough in the transformation of emotional corpus information. But there is still a gap between the real target and the synthesis speech. In order to narrow this gap, we propose an emotion speech synthesis method combining multi-channel Time–frequency Domain Generative Adversarial Networks (MC-TFD GANs) and Mixup. It includes three stages: multichannel Time–frequency Domain GANs (MC-TFD GANs), loss estimation based on Mixup and effective emotion region stacking based on Mixup. Among them, the gating unit GTLU (gated tanh linear units) and the image expression method of speech saliency region are designed. It combines the Time–frequency Domain MaskCycleGAN based on improved GTLU and the time-domain CycleGAN based on saliency region to form the multi-channel GAN in the first stage. Based on Mixup method, the calculation method of loss and the aggravation degree of emotion region are designed. Compared with several popular speech synthesis methods, the comparative experiments were carried out on the interactive emotional dynamic motion capture (IEMOCAP) corpus. The bi-directional three-layer long short-term memory (LSTM) model was used as the verification model. The experimental results showed that the mean opinion score (MOS) and the unweighted accuracy (UA) of the speech generated by the synthesis method were improved, and the improvements were 4% and 2.7%, respectively. The current model was superior to the existing GANs model in subjective evaluation and objective experiments, ensure that the speech generated by this model had higher reliability, better fluency and emotional expression ability.

This paper proposes an analyzable agent-based route choice modeling framework with good theoretical properties. This modeling framework allows heterogeneous individual learning rules and learning r...

This paper proposes a deep learning based multitask learning (MTL) model to predict network-wide traffic speed, and introduces two methods to improve the prediction performance. The nonlinear Granger causality analysis is used to detect the spatiotemporal causal relationship among various links so as to select the most informative features for the MTL model. Bayesian optimization is employed to tune the hyperparameters of the MTL model with limited computational costs. Numerical experiments are carried out with taxis’ GPS data in an urban road network of Changsha, China, and some conclusions are drawn as follows. The deep learning based MTL model outperforms four deep learning based single task learning (STL) models (i.e., Gated Recurrent Units network, Long Short-term Memory network, Convolutional Gated Recurrent Units network and Temporal Convolutional Network) and three other classic models (i.e., Support Vector Machine, k-Nearest Neighbors and Evolving Fuzzy Neural Network). The nonlinear Granger causality test provides a reliable guide to select the informative features from network-wide links for the MTL model. Compared with two other optimization approaches (i.e., grid search and random search), Bayesian optimization yields a better tuning performance for the MTL model in the prediction accuracy under the budgeted computation cost. In summary, the deep learning based MTL model with nonlinear Granger causality analysis and Bayesian optimization promises the accurate and efficient traffic speed prediction for a large-scale network.

Autonomous intersection management (AIM) refers to planning cooperative trajectories for multiple connected and automated vehicles (CAVs) when they pass through an unsignalized intersection. In modeling a generic AIM scheme, the predominant network-level or lane-level methods limit the cooperation potentiality of a multi-CAV team because 1) lane changes are forbidden or only allowed at discrete spots in the intersection, 2) each CAVs travel path is fixed or selected among a few topological choices, and 3) each CAVs travel velocity is fixed or set to a specified pattern. To overcome these limitations, this work models the intersection as a continuous free space and describes an AIM scheme as a multi-CAV trajectory optimization problem. Concretely, a centralized optimal control problem (OCP) is formulated and then numerically solved. To derive a satisfactory initial guess for the numerical optimization, a priority-based decentralized framework is proposed, wherein an x-y-time A* algorithm is adopted to generate a coarse trajectory for each CAV. To facilitate the OCP solution process, 1) the collision-avoidance constraints in the OCP are convexified, and 2) a stepwise computation strategy is adopted. Simulation results show the efficacy of the proposed offline AIM method.

A superpixel segmentation algorithm called multi-density peaks clustering (MDPC) is proposed. By selecting a sufficient number of local density maximum pixels from the image as cluster centers to depict the image texture, the boundary of the object can be captured very accurately. The algorithm framework of MDPC is divided into three steps. First, the local density of pixel is defined, and the local density maximum pixels are calculated. Then, the local density maximum pixels are used as cluster centers, and the global optimal search, which is based on the path-to-point idea, is used to complete the clustering of the remaining non-cluster center pixels to realize the initial segmentation. Finally, superpixels are obtained by merging the initial segments according to the size of the segments and the distance between adjacent segments. In quantitative comparisons, MDPC is compared with 13 state-of-the-art superpixel segmentation algorithms in three image segmentation datasets. The experimental results show that MDPC achieves better performance in terms of boundary recall, boundary precision, achievable segmentation accuracy, undersegmentation error, and explained variation. And the qualitative comparisons show that the proposed algorithm has obvious advantages over other superpixel segmentation algorithms in image detail description and boundary adherence. Finally, the practicability and stability of MDPC are further demonstrated by the application of image segmentation. The source code of MDPC will be available at https://github.com/zhaojianaaa.

In the mobile Internet era, demographic information plays an important role in many personalized services and research areas such as content recommendation [1], advertising [2], and behavioral prediction [3]. However, this information is treated as private data and is difficult to access. Therefore, the prediction of demographic information has aroused research interest. There are already some studies that predict demographic information based on user-generated content [4], smartphone sensor data, and user biometric data [5]. In this paper, we introduce a new method that uses App usage record data to predict the demographic information of smartphone users. App usage record data is a set of easily accessible, rich and insightful data. We propose a deep learning framework that uses a convolutional neural network(CNN) to predict the demographic information of smartphone users through app usage record data. We use a set of data with 6-months App usage record of about 5,000 users to train our prediction model. The CNN model we proposed has achieved good prediction performance on the training data set. In the comparative experiment, the prediction effect of our model is also better than the effect of the three classical machine learning algorithms.

In recent years, the application of style transfer has become more and more widespread. Traditional deep learning-based style transfer networks often have problems such as image distortion, loss of detailed information, partial content disappearance, and transfer errors. The style transfer network based on deep learning that we propose in this article is aimed at dealing with these problems. Our method uses image edge information fusion and semantic segmentation technology to constrain the image structure before and after the migration, so that the converted image maintains structural consistency and integrity. We have verified that this method can successfully suppress image conversion distortion in most scenarios, and can generate good results.

Knowledge of trip travel times serves an important role in transportation management and control. Existing travel time estimation approaches generally cover empirical ones, statistical ones and hybrid ones. Despite strong tractability, the empirical approaches cannot sufficiently capture diverse travel time distributions (TTDs) and often encounter some issues (e.g., assumption of a predefined distribution, failure of significance tests). Statistical and hybrid methods possess better generalization in estimating heterogeneous TTDs, but fail to model the network-wide spatiotemporal correlations, which have been found useful in the TTD estimation. To address these drawbacks, this paper proposes a deep learning based Trip Information Maximizing Generative Adversarial Network (T-InfoGAN). In this method, the trip TTD is estimated by modeling the joint distribution of travel times of two successive links with the consideration of network-wide spatiotemporal correlations. Meanwhile, a dynamic clustering with Wasserstein distance (DCWD) algorithm is used to explore the traffic state transitions for link pairs and cluster the link pairs with similar TTDs into one group, which benefits the training and estimation processes of T-InfoGAN. Then, based on GPS trajectory data from Didi Chuxing in Chengdu city, China, numerical results show that the T-InfoGAN with DCWD can well estimate three mini trip TTDs with various features, and performs better than three other counterparts (i.e., Convolution method, MC-Grid method, and MC-GMMS method) in estimating the TTDs of two longer trips. In summary, this study is the first successful try to estimate trip TTDs within the framework of Generative Adversarial Networks (GANs), and the deep learning based T-InfoGAN is a promising approach to estimate heterogeneous trip TTDs with the better generalization and flexibility in the big data era.

Credit scoring represents a two-classification problem. Moreover, the data imbalance of the credit data sets, where one class contains a small number of data samples and the other contains a large number of data samples, is an often problem. Therefore, if only a traditional classifier is used to classify the data, the final classification effect will be affected. To improve the classification of the credit data sets, a Gaussian mixture model based combined resampling algorithm is proposed. This resampling approach first determines the number of samples of the majority class and the minority class using a sampling factor. Then, the Gaussian mixture clustering is used for undersampling of the majority of samples, and the synthetic minority oversampling technique is used for the rest of the samples, so an eventual imbalance problem is eliminated. Here we compare several resampling methods commonly used in the analysis of imbalanced credit data sets. The obtained experimental results demonstrate that the proposed method consistently improves classification performances such as F-measure, AUC, G-mean, and so on. In addition, the method has strong robustness for credit data sets.

Extensive empirical evidence indicates that travellers consider a number of qualities (e.g. travel time, monetary cost, and reliability) when deciding between alternative routes. This study focused on monetary cost and travel time and reviewed two traditional user equilibrium models that incorporate both factors: the VOT-based generalised cost user equilibrium (GCUE) and the bi-objective user equilibrium (BUE). Several properties and assumptions of these models are highlighted that may not be realistic. The present paper develops a bi-attribute regret-minimisation user equilibrium (BRminUE) model, in which travellers aim to minimise their regret rather than maximizing their utility in their travel choices. The relationships between GCUE, BRminUE, and BUE were investigated and a simple example verifies the proved properties. The BRminUE model is a special case of the BUE model; however, it is more general than the GCUE model. Numerical analyses on a four-node tolled network indicate the performance of the ...

Lane change is a basic and critical element of complicated driving maneuvers such as overtaking, merging, and exit. Improper lane change is a primary cause for car crashes. This article foc...

An input vector composed of various features plays an important role in short-term traffic forecasting. However, there is limited research on the optimal feature selection of an input vector for a certain forecasting task. To fill the gap, this paper proposes a cohesion-based heuristic feature selection method by analyzing the nature of the forecasting methods. This method is able to determine which features should be contained in an input vector to make a forecasting algorithm perform better. The proposed method is demonstrated in two experiments based on the empirical traffic flow data. The results show that the method is able to improve the performances of the short-term traffic forecasting algorithms. It is then suggested to consider the proposed method as a preprocessing procedure in practical forecasting applications.

The incidence rate of depression, as one of the most prevalent mental disorders, is the biggest challenge in the process of diagnosing mental illness. This paper studies the application of emotion recognition in patients with depression based on speech signal. In order to improve the recognition accuracy of depression, usually, we can get better recognition effect from two aspects: extracting the representative speech features of patients with depression and using different classification methods. In this paper, by extracting the features of speech speed, short-term average energy, gene frequency, Mel frequency cepstrum coefficient, we combine generative adversarial network and convolutional neural network to study the emotion recognition of depression. Experimental results show that the proposed method achieves the best recognition effect on AViD-Corp dataset.

In order to design a programmable multipath waveform generator, VHDL is used to design every module. It is consisted with the waveform input, FPGA module, digital-to-analog conversion (D/A) module, display circuit. By pressing the keys, eight waves can be chosen and shown in the oscilloscope. With the characteristics of low cost, portable and expandable, it can be widely used in electronic circuit system.

In order to solve the reality of the traditional manual operation and further improve the accuracy and efficiency of the flag raising, an automatic flag lifting system is designed based on the single-chip embedded technology. It is consisted with the single-chip micro-controller minimum system, motor circuit, key circuit, LCD display circuit, voice control circuit, DC circuit. By pressing the button, the national flag begin to rise and fall with playing the national anthem at the same time. The system has the advantages of low price, high stability and simple operation.

In recent years, consumer credit has flourished in China. A credit factory is an important mode to speed up the loan application process. Order scheduling in credit factories belongs to the np-hard problem and it has great significance for credit factory efficiency. In this work, we formulate order scheduling in credit factories as a multi-agent reinforcement learning (MARL) task. In the proposed MARL algorithm, we explore a new reward mechanism, including reward calculation and reward assignment, which is suitable for this task. Moreover, we use a convolutional auto-encoder to generate multi-agent state. To avoid physical costs during MARL training, we establish a simulator, named Virtual Credit Factory, to pre-train the MARL algorithm. Through experiments in Virtual Credit Factory and an A/B test in a real application, we compare the performance of the proposed MARL approach and some classic heuristic approaches. In both cases, the results demonstrate that the MARL approach has better performance and strong robustness.

Nowadays, speech corpus plays a fundamental role in the research and development of speech processing technology. This paper mainly focuses on the research and analysis of adult emotional speech corpus construction methods. According to the needs of the research on the emotion recognition of reading speech and spoken speech, this paper studies the construction standard and the corresponding annotation standard of the emotion speech recognition for the natural environment, and designs the specific scheme for the evaluation of the effectiveness of the corpus. Based on this, an emotional speech corpus with multi-level annotation information is constructed. Experiments show that the corpus has balanced coverage of local and global emotional expression while retaining the natural attributes of emotion, which provides a reliable data support for the research of emotion based on speech recognition technology.

We show that correlations between the camera used to acquire an image and the class label of that image can be exploited by convolutional neural networks (CNN), resulting in a model that “cheats” at an image classification task by recognizing which camera took the image and inferring the class label from the camera. We show that models trained on a dataset with camera / label correlations do not generalize well to images in which those correlations are absent, nor to images from unencountered cameras. Furthermore, we investigate which visual features they are exploiting for camera recognition. Our experiments present evidence against the importance of global color statistics, lens deformation and chromatic aberration, and in favor of high frequency features, which may be introduced by image processing algorithms built into the cameras.

In order to solve the cascading failure problem of real networks, it is important to study the robustness of coupled networks. In this paper, a multi-subnet complex complex network model is used to build a cascading failure model on the coupled network. The influence of network size, adjustable parameters, and coupling factors on the robustness of the coupled network is mainly studied. Research shows that when both subnets are BA networks, the greater the contribution of the coupling relationship to the initial load of the nodes, and adopt heterogeneous coupling, networks will more robust.

The resolution of an image is an important indicator for measuring image quality. The higher the resolution, the more detailed information is contained in the image, which is more conducive to subsequent image analysis and other tasks. Improving the resolution of images has always been the unremitting pursuit of industry and academia. In the past, people used hardware devices to increase the resolution, which is a practical solution. However, there are many limitations in the method of improving the image resolution by hardware devices. We use software-based image super-resolution technology, which transforms low-resolution images into high-resolution images through a series of machine learning algorithms. The classic GAN algorithm is difficult to train a model, and the improved Wasserstein GAN algorithm can make the model training more stable. Based on SRGAN model, this algorithm replaces the classical GAN algorithm with the improved WGAN algorithm. We will use the FY-3D satellite’s Medium Resolution Spectral Imager Type II (MERSI-II) data, using super-resolution algorithms to make the reconstructed image significantly better visually. We conducted four sets of controlled experiments using four different improved methods. We will evaluate the image from three aspects: peak signal to noise ratio value, structural similarity value and visual effect. We applied the WGAN-GP algorithm to super-resolution tasks and achieved the desired results.

Speech emotion recognition is a challenging and widely examined research topic in the field of speech processing. The accuracy of existing models in speech emotion recognition tasks is not high, and the generalization ability is not strong. Since the feature set and model design of effective speech directly affect the accuracy of speech emotion recognition, research on features and models is important. Because emotional expression is often correlated with the global features, local features, and model design of speech, it is often difficult to find a universal solution for effective speech emotion recognition. Based on this, the main research purpose of this paper is to generate general emotion features in speech signals from different angles, and use the ensemble learning model to perform emotion recognition tasks. It is divided into the following aspects: (1) Three expert roles of speech emotion recognition are designed. Expert 1 focuses on three-dimensional feature extraction of local signals, expert 2 focuses on extraction of comprehensive information in local data, and expert 3 emphasizes global features: acoustic feature descriptors (low-level descriptors (LLDs)), high-level statistics functionals (HSFs), and local features and their timing relationships. A single-/multiple-level deep learning model that meets expert characteristics is designed for each expert, including convolutional neural network (CNN), bi-directional long short-term memory (BLSTM), and gated recurrent unit (GRU). Convolutional recurrent neural network (CRNN), based on a combination of an attention mechanism, is used for internal training of experts. (2) By designing an ensemble learning model, each expert can play to its own advantages and evaluate speech emotions from different focuses. (3) Through experiments, the performance of various experts and ensemble learning models in emotion recognition is compared in the Interactive Emotional Dyadic Motion Capture (IEMOCAP) corpus and the validity of the proposed model is verified.

Deep Convolutional Neural Networks (CNN) have been extensively applied in various computer vision tasks. Although such approaches have demonstrated exceptionally high performance in various open challenges, adapting them to more specialised tasks can be non-trivial. In this paper we discuss our design and implementation of a batchcode detection system capable of accurate segmentation of batchcode regions within images of consumer products. A batchcode is a unique identifier printed on the packaging of many products that encodes useful information such as date and location of manufacture. Detection of batchcodes in images of products is a useful step in many processes, including quality control, supply chain tracking and counterfeit detection. Beginning with a unique dataset of product images and a set of crowdsourced coarse annotations that roughly correspond to the locations of batchcodes, we demonstrate that such annotations are insufficient for training a reliable model, and subsequently describe a novel label refinement process, which we call the Maximally Stable Global Region (MSGR) method, that we use to generate accurate ground-truth data suitable for training a robust neural network. We also show that detection accuracy can be further improved by applying MSGR to the output of the neural network. We evaluate our approach using a manually labelled test dataset of images of shampoo bottles, and demonstrate the efficacy of the proposed method for accurate real-time batchcode detection.

The authors present an improved feature selection solution for the view-invariant gait recognition problem, based on their previously proposed method called view-invariant feature selector (ViFS), which automatically reconstruct an optimised gallery template from a set of multi-view gallery templates. They improved ViFS by introducing a constraint to make sure that the reconstructed features have the same scale as the original features, thus reducing the number of misclassifications caused by data misalignment. They evaluate the improved ViFS on the CASIA B and OU-ISIR large population datasets by performing a wide range of comparative studies in order to explore and confirm its effectiveness. Evaluation results indicate that the proposed framework is very effective for view-invariant gait recognition tasks.

The average is usually applied to describe the traffic delay, and the environmental factors are considered to influence it. However, the distribution and the influencing individual behavior on traf...

The variation of the electrical distance and the complexity of the electric network lead to the variations of feeder impedances between distributed generation units and load points. It is determined that conventional droop control has drawbacks in achieving accurate power sharing due to the effects of mismatched impedance. Therefore, this study proposes an enhanced proportional power sharing strategy based on adaptive virtual impedance in a low-voltage networked microgrid. The improved R-L type droop control can effectively prevent the coupling between real and reactive powers. Furthermore, an adaptive virtual impedance loop is introduced to counteract the feeder voltage drop. The method utilises real and reactive power mismatching which were fed into integral controllers, and then generates the virtual inductive and resistive components, respectively. This proposed strategy is able to enhance power sharing accuracy without requiring the knowledge of feeder impedance, and it is more adaptive to the complex impedance. The simulation experiments carried out under the environment of MATLAB/Simulink, and results verify the effectiveness of the proposed strategy.

This paper studies the traffic flow in a grid road network with non-signalized intersections. The nature of the drivers in the network is simulated such that they play an iterative snowdrift game with other drivers. A cellular automata model is applied to study the characteristics of the traffic flow and the evolution of the behaviour of the drivers during the game. The drivers use best-response as their strategy to update rules. Three major findings are revealed. First, the cooperation rate in simulation experiences staircase-shaped drop as cost to benefit ratio r increases, and cooperation rate can be derived analytically as a function of cost to benefit ratio r . Second, we find that higher cooperation rate corresponds to higher average speed, lower density and higher flow. This reveals that defectors deteriorate the efficiency of traffic on non-signalized intersections. Third, the system experiences more randomness when the density is low because the drivers will not have much opportunity to update strategy when the density is low. These findings help to show how the strategy of drivers in a traffic network evolves and how their interactions influence the overall performance of the traffic system.

Credit scoring has been attracting increasing attention in the Chinese consumer financial industry. Traditional approaches are easily influenced by sample selection bias because they use accepted applicant samples only, while the applicant population also includes rejected applicants. Reject inference is a technique to infer good/bad labels for rejected applicants, which can overcome biases in credit scoring. However, previously proposed reject inference methods usually ignore the imbalanced distribution in accepted data, which means that good applicants are much more than bad ones in most practical consumer loan applications. Both the neglect of rejected data and the imbalanced distribution in accepted data weaken the performance of current credit scoring models. In this paper, we propose a novel reject inference framework that takes into account the imbalanced data distribution for consumer credit scoring. First, we use an advanced graph-based semi-supervised learning algorithm to solve the reject inference problem, which is called label spreading. Second, faced with an imbalanced distribution of good and bad samples in accepted applicants, we conduct imbalanced learning using a modified Synthetic Minority Over-sampling Technique before reject inference. Then, six binary classifiers are studied in our proposed framework for credit scoring modeling. Finally, we present the results of four exact experiments as well as online A/B tests for performance evaluation using data provided by a leading Chinese fintech company. Empirical results indicate that the proposed framework performs better than traditional scoring models across different evaluation metrics, representing a progressive method that promotes credit scoring research as well as improving fintech practices.

Multi-vehicle motion planning (MVMP) refers to computing feasible trajectories for multiple vehicles. MVMP problems are generally solved in two ways, namely simultaneous methods and joint methods. An inherent difference between both types of methods is that, simultaneous methods compute motions for vehicles all at once, while joint methods divide the original problem into parts and combine them together. The joint methods usually sacrifice solution quality for computational efficiency, and the simultaneous methods are applicable to simple or simplified scenarios only. These defects motivate us to develop an efficient simultaneous computation method which provides high-quality solutions in generic cases. Progressively constrained dynamic optimization (PCDO), an initialization-based computation framework is proposed to ease the burdens of simultaneous computation methodologies when they are adopted to solve the MVMP problems. Specifically, PCDO locates and discards the redundant constraints in the MVMP problem formulation so as to reduce the problem scale, thereby easing the problem-solving process. Our simulations focus on the cooperative parking scheme of automated vehicles. Comparative simulation results show that (1) the designs in PCDO are efficient, and (2) simultaneous computation outperforms joint computation.

We present a method for identifying bus rapid transit routes optimized for serving the greatest number of passengers. Because formulating a relevant model is a complex problem, a particle swarm opt...

Social interaction is increasingly recognized as an important factor that influences travelers’ behaviors. It remains challenging to incorporate its effect into travel choice behaviors, although there has been some research into this area. Considering random interaction among travelers, we model travelers’ day-to-day route choice under the uncertain traffic condition. We further explore the evolution of network flow based on the individual-level route choice model, though that travelers are heterogeneous in decision-making under the random-interaction scheme. We analyze and prove the existence of equilibrium and the stability of equilibrium. We also analyzed and described the specific properties of the network flow evolution and travelers’ behaviors. Two interesting phenomena are found in this study. First, the number of travelers that an individual interacts with can affect his route choice strategy. However, the interaction count exerts no influence on the evolution of network flow at the aggregate-level. Second, when the network flow reaches equilibrium, the route choice strategy at the individual-level is not necessarily invariable. Finally, two networks are used as numerical examples to show model properties and to demonstrate the two study phenomena. This study improves the understanding of travelers’ route choice dynamics and informs how the network flow evolves under the influence of social interaction.

Recent studies on multi-label image classification have focused on designing more complex architectures of deep neural networks such as the use of attention mechanisms and region proposal networks. Although performance gains have been reported, the backbone deep models of the proposed approaches and the evaluation metrics employed in different works vary, making it difficult to compare each fairly. Moreover, due to the lack of properly investigated baselines, the advantage introduced by the proposed techniques are often ambiguous. To address these issues, we make a thorough investigation of the mainstream deep convolutional neural network architectures for multi-label image classification and present a strong baseline. With the use of proper data augmentation techniques and model ensembles, the basic deep architectures can achieve better performance than many existing more complex ones on three benchmark datasets, providing great insight for the future studies on multi-label image classification., Comment: IEEE International Conference on Image Processing 2019

Most modern people ignore the importance of reading aloud. However, for children aged 5-12, reading aloud is not only an essential skill in the learning process, but also an effective means of cultivating sentiment. Because there is a nonlinear relationship between the characteristics of the spoken speech signal and the evaluation criteria, the emotional features suitable for children's reading evaluation are extracted from the audio signal, which is very important for the recognition of children's reading emotions. However, automatically recognizing emotions from speech is a challenging task, and its recognition depends on the validity of the speech emotion features and the accuracy of the model. In this research, we start with traditional Low Level Descriptors (LLD) to learn emotion-related features automatically which were found in speech, using High Level Statistics Functions (HSF), and emotion-related Short time frame level acoustic features can be learned. These features are appropriately aggregated into a compact feature representation in conjunction with a spectrogram to form a set of features that effectively characterize the emotion signal. The proposed solution is evaluated on the children's emotional reading speech library and shows more accurate predictions than existing emotion recognition algorithms.

In order to adapt to the regulation and management requirements of the integrated energy system, this paper carried out the research of integrated energy intelligent management and control system. According to the design concept of plantwide control, the hierarchical structure of multi-energy complementary coordinated control, energy regulation and management was proposed. In the multi-energy complementary coordinated control, a unified coordination controller is designed to comprehensively coordinate various control methods to ensure the safe and stable operation of the regional integrated energy system. The energy regulation and management system is inherited, and the information exchange with the energy trading and decision-making system is carried out. It has the functions of the energy efficiency analysis, energy monitoring, energy scheduling and optimization, "generation-grid-load-storage" collaborative optimization, which could provide optimized settings for the lower layer control, and achieve regional integrated energy stratification and control, and improve the whole energy efficiency of the energy network.

Along with the booming consumer credit market, credit scoring has received an increasing concern in auto financial companies. However, the modeling without rejected applicants and the imbalanced distribution of accepted examples affect the predictive performance. In this paper, we propose a novel framework for credit scoring using an imbalanced-learning-based reject inference. First, we employ an imbalanced learning for the accepted applicant data using Synthetic Minority Over-sampling Technique for reject inference. Second, we conduct reject inference for rejected applicants based on a graph-based semi-supervised learning algorithm, which is called label propagation. Third, we use tree-based ensemble learning models as base classifiers to train the combined training data. Finally, we give an exact experiment for assessment using data from a Chinese auto loan company. The results indicate that the proposed novel framework performs better than comparative models, which represents a progressive method for auto loan.

The development of urban drainage systems is challenged by rapid urbanization, however, little attention is paid to the urban form and its effects on these systems. This study develops an integrated city-drainage model that configures typical urban forms and their associated drainage infrastructures, specifically domestic wastewater and rainwater systems, to analyze the relationship between them. Three typical types of urban forms were investigated: the square, the star, and the strip. Virtual cities were designed first, with the corresponding drainage systems generated automatically and then linked to a model herein called the Storm Water Management Model (SWMM). Evaluation was based on 200 random configurations of wastewater/rainwater systems with different structures or attributes. The results show that urban forms play more important roles on three dimensions of performance, namely economic efficiency, effectiveness, and adaptability, of the rainwater systems than of the wastewater systems. Cost is positively correlated to the effectiveness of rainwater systems among the different urban forms, while adaptability is negatively correlated to the other two performance dimensions. Regardless of the form, it is difficult for a city to make its drainage systems simultaneously cost-effective, efficient, and adaptable based on the virtual cities we investigated. This study could inspire the urban planning of both built-up and to-be-built areas to become more sustainable with their drainage infrastructure by recognizing the pros and cons of different macroscale urban forms.

In life, the voice signals collected by people are essentially mixed signals, which mainly include information related to speaker characteristics, such as gender, age and emotional state. The commonality and characteristics of traditional single-dimensional speaker information recognition are analyzed, and children’s individualized analysis is carried out for common acoustic feature parameters such as prosodic features, sound quality features and spectral-based features. Therefore, considering the temporal characteristics of voice, combined with the Time-Delay Neural Network (TDNN) model, Bidirectional Long Short-Term Memory model and the attention mechanism, the multi-channel model is trained to form a speaker recognition problem solution for children’s speaker recognition. A large number of experimental results show that on the basis of guaranteeing the accuracy of age and gender recognition, higher accuracy of children’s voiceprint recognition can be obtained.

In the process of human social activities and daily communication, speech, text and facial expressions are considered as the main channels to convey human emotions. In this paper, a fusion method of multi-modal emotion recognition based on speech, text and motion is proposed. In the speech emotion recognition (SER), a depth wavefield extrapolation - improved wave physics model (DWE-WPM) is designed. In order to simulate the information mining process of LSTM, a user-defined feature extraction scheme is used to reconstruct the wave and inject it into DWE-WPM. In the text emotion recognition (TER), the transformer model with multi attention mechanism is used to recognize the text emotion combined. In the motion emotion recognition (MER), the sequential features of facial expression and hand action are extracted in groups. Combined with the bidirectional three-layer LSTM model with attention mechanism, a joint model of four channels is designed. Experimental results show that the proposed method has high recognition accuracy in multi-modal, and the accuracy is improved by 9% in the interactive emotional dynamic motion capture (IEMOCAP) corpus.

Coastal aquaculture areas are some of the main areas to obtain marine fishery resources and are vulnerable to storm-tide disasters. Obtaining the information of coastal aquaculture areas quickly and accurately is important for the scientific management and planning of aquaculture resources. Recently, deep neural networks have been widely used in remote sensing to deal with many problems, such as scene classification and object detection, and there are many data sources with different spatial resolutions and different uses with the development of remote sensing technology. Thus, using deep learning networks to extract coastal aquaculture areas often encounters the following problems: (1) the difficulty in labeling; (2) the poor robustness of the model; (3) the spatial resolution of the image to be processed is inconsistent with that of the existing samples. In order to fix these problems, this paper proposes a novel semi-/weakly-supervised method, the semi-/weakly-supervised semantic segmentation network (Semi-SSN), and adopts 3 data sources: GaoFen-2 image, GaoFen-1(PMS)image, and GanFen-1(WFV)image with a 0.8 m, 2 m, and 16 m spatial resolution, respectively, and through experiments, we analyze the extraction effect of the model comprehensively. After comparing with other the-state-of-art methods and verifying on an open remote sensing dataset, we take the Fujian coastal area (mainly Sanduo) as the experimental area and employ our method to detect the effect of storm-tide disasters on coastal aquaculture areas, monitor the production, and make the distribution map of coastal aquaculture areas.

The generation of emotional speech is a challenging and widely applied research topic in the field of speech processing. Because the design method of effective speech feature expression and generation model directly affects the accuracy of emotional speech generation, it is difficult to find a general solution of emotional speech synthesis. In this paper, the CycleGAN model is used as the starting point, and the improved convolution neural network (CNN) model and identity mapping loss scheme are used to achieve effective timing information capture. At the same time, we learn the positive mapping and the reverse mapping to find the best matching design scheme, and retain the speech information in this process, without relying on other audio data. Experiments show that the emotional speech can be accurately recognized by comparing the speech emotion before and after the improvement on the speech corpus of children’s reading. By comparing with the common emotional speech generation model, the advantages of the model proposed in this paper are verified.

According to the characteristics of too many types of plane parameter, increasing parameters needing to be collection, complex system and diverse requirements of aircraft flight data acquisition, a configurable aircraft data acquisition system based on FPGA + CPU was designed. This system architecture can give full play to the parallel working ability of FPGA in which to collecting multi-type and multichannel flight data simultaneously. At the same time, the sequential function of CPU is used to realize system configuration loading conveniently and complete flight data acquisition tasks without complex operation of professional technicians on the plane. This system has the advantages of simple circuit structure, high integration and high universal, which can be used for engine sensor parameter, air frame sensor parameter, flight attitude and other flight data.

Gait recognition is one of the most important techniques for human identification at a distance. Most current gait recognition frameworks consist of several separate steps: silhouette segmentation, feature extraction, feature learning, and similarity measurement. These modules are mutually independent with each part fixed, resulting in a suboptimal performance in challenging conditions. In this paper, we integrate those steps into one framework, i.e., an end-to-end network for gait recognition, named GaitNet. It is composed of two convolutional neural networks: one corresponds to gait segmentation, and the other corresponds to classification. The two networks are modeled in one joint learning procedure which can be trained jointly. This strategy greatly simplifies the traditional step-by-step manner and is thus much more efficient for practical applications. Moreover, joint learning can automatically adjust each part to fit the global optimal objective, leading to obvious performance improvement over separate learning. We evaluate our method on three large scale gait datasets, including CASIA-B, SZU RGB-D Gait and a newly built database with complex dynamic outdoor backgrounds. Extensive experimental results show that the proposed method is effective and achieves the state-of-the-art results. The code and data will be released upon request.

Intelligent medical interaction system is a platform based on Internet of things and cloud platform, using radio frequency identification technology, wireless communication technology, network communication technology and cloud storage technology. As a new operation mode of medical management communication system that allows patients, doctors and family members to have seamless connection, it achieves the real sense of patient intelligence and communication between doctors and patients. This system develops identity recognition, sample recognition, medical record recognition and other functions. It can get online information on doctors and patients, and store them in the cloud platform. At the same time, it can achieve timely communication between doctors, patients and their families. The system consists of network layer, control layer and terminal application system, including mobile nursing, disinfection tracking, medical waste management, infant anti-theft, tele consultation, pathological examination and remote health education subsystem. Through this system, the purpose of strengthening the connection between the hospital and the patients can be achieved.

This paper focuses on the scheme of cooperative lane change motion planning of multiple connected and automated vehicles, so as to minimize the time for lane change while penalizing large steering angles subject to hard collision avoidance constraints. Nominally this scheme should be formulated in a centralized way with the constraints of all the vehicles considered simultaneously. In order to facilitate the numerical solving process of this centralized optimization problem, we propose a stepwise computation framework. Starting with a sub-problem with all of the collision avoidance constraints removed, a sequence of sub-problems are defined by adding back the removed collision avoidance constraints gradually until the original problem takes shape in the end. The optimum of one sub-problem is always used as the initial guess when solving the next sub-problem. This iterative process continues until the optimum of the original problem is obtained. In this way, the difficulties in the original centralized problem are divided into multiple parts, and every progress made to address the partial difficulties is “solidified” by the initial guess.

In this paper, we propose a cooperative motion planning method for a group of connected and automated vehicles (CAVs) crossing a lane-free intersection without using explicit traffic signaling. This multi-vehicle motion planning task is formulated as a centralized optimal control problem. However, the solution to this optimal control problem is numerically intractable due to the high dimensionality of the collision-avoidance constraints and the nonlinearity of the vehicle kinematics. A two-stage strategy is proposed for generating online solutions: at Stage 1, the CAVs are requested to reach a standard formation before entering the intersection; at Stage 2, the vehicles cross the intersection. As the motion planning sub-problem at Stage 2 begins with a standard configuration, the optimal solution to this standard sub-problem can be computed offline in advance and applied online directly. On the other hand, the formation reconfiguration sub-problem at Stage 1 is easy to solve online. Through dividing the entire dynamic process into two periods, the difficulties in the original optimal control problem are significantly reduced so that the real-time performance is achieved.

Passengers’ boarding and alighting is an important link between station and the train in urban rail transit system. The paper presents features of boarding and alighting of passenger flow. A model was established with two continuity equations and a formula. In that model, continuity equations about boarding and alighting are deduced based on these two characteristics, revealing the relationship between time and density. The formula of resultant about passenger flow during their boarding and alighting activities is aimed at exposing the main direction of the whole procedure. What’s more, the model was applied to Chongwenmen station in Beijing subway Line 1, the model is simplified for simulation by a mathematical application software-MATLAB, graphing directly. The result is formally described and experimented in experimental and real-world situation, which is proved to be correct and reliable. The model provides the theory foundation for safer process of boarding and alighting for actual operation and plays an important guiding meaning in daily management.

To better capture typical individual response modes under strategic uncertainty in congestible networks, we conducted laboratory experiments in a network with two parallel routes under within-subject design. Sixty-four undergraduates were assigned into four sessions to make recurrent route-choice decisions under Condition partial-information (PI) first, and then under Condition full-information (FI). Individuals whose response modes are featured by a series of conditional probabilities regarding switching behaviour naturally cluster into three and four groups under Conditions PI and FI, respectively. An in-depth analysis of behavioural bases of each type was discussed. In Condition FI, the proportion of highly responsive players (holding Direct-response-like and Contrary-response-like patterns) and Highly-risk-averse players drops, whereas the Status-quo-maintenance category players stand out. More feedback information was disclosed for the purpose of reducing uncertainty but turned out to reduce the proportion of people who were highly responsive to the new information and who firmly commit themselves to a unique route.