Abstract:With the discovery, management and application of data associations, graph databases develop rapidly. By analyzing the concept, model and structure of graph databases, this study summarizes the characteristics of graph databases. Regarding graph databases, the study expounds their three key technologies, compares existing products, and summarizes their current application scenarios. Finally, it proposes the trend of graph database development in the future.

Abstract:In the Internet of Vehicles (IoV) scenario of 5G mobile edge computing (MEC), a server selection scheme based on task priority is designed for the selection of vehicle task offloading targets. Considering the influence of time, energy consumption, costs, and other factors on the choice of offloading locations, a solution based on the multi-index auction game is proposed. Through the multi-index auction mechanism, the optimal MEC server is selected to provide task offloading services for vehicles, which realizes the Bayesian Nash equilibrium of the cooperation between vehicles and RSU. Simulations show that this scheme can reduce the total cost of task offloading and meet multiple performance indicators under the constraints of guaranteeing vehicle task offloading time and energy consumption.

Abstract:As for contrast-enhanced ultrasound video cases involving cervical lymph nodes, perfusion features can be extracted from the time-intensity curve (TIC) to make a diagnosis. The existing research methods that analyze a region of interest at the pixel level are able to describe perfusion features accurately. However, in-depth research on methods of visualizing the perfusion flow direction is limited in number. In this study, pixel-based TIC analysis is performed, and the TIC is screened by double screening. Then, two-dimensional perfusion parameters are extracted from the TIC to visualize the perfusion flow direction. The streamline image generated after feature extraction can reflect the distribution of blood vessels to a certain extent, thereby helping doctors with their diagnoses and providing implications to microvascular reconstruction.

Abstract:With the rapid development of network communication technology and applications, more refined and differentiated data transmission requirements of applications are put forward. However, the low flexibility of traditional network transmission protocols makes it unable to meet the differentiated requirements of various applications, and thus there is an urgent need to study differentiated reliable transmission control protocols for future scenarios. This study proposes a differentiated reliable transmission protocol and focuses on a reliability-based differentiated reliable transmission congestion control mechanism. Different congestion avoidance and congestion recovery strategies are designed for different reliability differences, and the bandwidth estimation strategy is employed to accurately adjust the congestion threshold. Experiments verify that the differentiated reliable transmission congestion mechanism has greater improvement in transmission efficiency than the packet loss and delay-based congestion algorithms and at the same time can achieve good fairness.

Abstract:Medical image registration plays a crucial role in medical image processing and analysis. Due to the large differences in gray scale and texture information between quantitative susceptibility mapping (QSM) and T1-weighted images, it is difficult for existing medical image registration algorithms to obtain accurate registration results efficiently. Therefore, this study proposes an unsupervised deep learning registration model (residual fusion registration network, RF-RegNet) based on residual fusion. RF-RegNet is composed of an encoder, a decoder, a resampler, and a context-similarity feature extractor. The encoder and decoder are used to extract the features of the image pair to be aligned and estimate the displacement vector field (DVF). The moving QSM image is resampled according to the estimated DVF, and the context-similarity feature extractor is used to extract separately the context-similarity features of the reference T1-weighted image and the resampled QSM image to describe the similarity of the two images. The mean absolute error (MAE) between context-similarity features from the two images is used to drive the convolutional neural network (ConvNet) learning. Experimental results reveal that the proposed method significantly improves the registration accuracy of QSM images and T1-weighted images, which is adequate for clinical demands.

Abstract:In view of the problems existing in the risk assessment of power monitoring systems, such as incomplete system modeling, fuzzy evaluation opinions of experts, and lack of consideration of the overall risk of systems, a risk assessment method for power monitoring systems is proposed, which is based on the cloud model and improved evidence theory. Firstly, according to the structure and security requirements of a power monitoring system, the equipment, security objectives, and threats of the system are analyzed, and the overall risk assessment model of the system is built. Then, in combination with the FAHP and modified entropy weight method, the weight of each element is obtained by using the optimal combination weighting method. Finally, the comprehensive risk assessment of the power monitoring system is completed by the cloud model and improved evidence theory, and the risk level of the system is obtained. The simulations show that the method is feasible and effective, which provides a new idea for the security management of the power monitoring system.

Abstract:With the continuous development of cloud computing, more and more cloud service providers use cloud platforms to provide services for cloud users. The configuration and implementation of cloud services are not clear to the cloud platform, resulting in a lack of trust. Various methods have been proposed in China and abroad to evaluate the credibility of cloud services in response to this problem. However, the objectivity of evaluation is less considered. This study proposes the environmental indicator, using the sliding window combined with subjective and objective evaluation to conduct credibility evaluation of cloud services. It proves through experiments that this evaluation method improves the accuracy of cloud service credibility evaluation.

Abstract:The development of the new-generation science and technology and interdisciplinary and cross-field integration have greatly improved the smartness of fields such as parks, transportation, energy, and security. However, due to the data barriers and lack of cross-business and cross-system collaboration, the smartness of city operation and management is still low. Taking the Happiness Forest Belt (XingFu LinDai) Project as an example, this study aims to fully and efficiently use the daily supervision and monitoring data on this forest belt to meet basic city operation and management requirements such as service, operation, management, visual display, and auxiliary emergency command decision-making. For this purpose, it designs a system architecture and a function system from the top down and then builds a smart operation and management platform that is expected to serve as a reference for smart city construction.

Abstract:Currently, water quality monitoring in China is faced with many problems, such as limited monitoring points, heavy communication costs, and high power consumption. To solve these problems, this study constructs a water quality monitoring system based on the LoRa wireless transmission technology. This system is composed of a data acquisition terminal, a wireless communication circuit, and a remote-control center. The embedded technology and the FreeRTOS system are used for sensor task scheduling and data collection. The processed data are then sent to the remote-control center through a wireless communication circuit with the LoRa technology and finally displayed in the upper computer software, which, implemented in Qt, can monitor sensor status and location of monitoring points in real time. The test results show that the coefficients of variation of the five data including pH are less than 5% and that of temperature is as low as 8.76%, which means the water quality of the Bahe River can be monitored stably and continuously for a period of time. With the advantages of low cost, strong expansibility, and high integration, the proposed system has good application prospects in water environments of sewage monitoring and aquaculture.

Abstract:Gait recognition is an emerging biometric technology, which can be widely used in criminal security, epidemic transmission chain tracking, etc. The essence of this technology is to identify people’s identity, age, gender and other biological attributes through their human body shape and walking posture. Compared with other biometric technologies, gait recognition has significant advantages such as long distance, full view, no perception, and anti-counterfeiting. In this study, we design a cross-view gait tracking system for multiple people and multiple biological attributes. The system fully considers the impact of covariates (such as multiple people, cross view and clothing change) on gait recognition accuracy in real application scenarios. It extracts the gait information of pedestrians from complex environments to accurately analyze their biological attributes such as identity, age, and gender through a more robust algorithm design. The experimental results show that the accuracy of the deep learning-based gait recognition algorithm model in this system can reach 88.0% in the case of cross view and multiple walking states and 94.8% in the case of multiple views for gender classification. The average age error of age estimation is about 7.92 years with a standard deviation of about 8.11. These results are better than those of recent algorithms in related fields and reach a relatively leading level. At a low development cost, the system is oriented to application scenarios and supports real-time gait detection.

Abstract:China’s transaction platform for grid procurement features a huge transaction amount, a large number of suppliers, and strict process control requirements. The transaction data platform for grid procurement proposed in this study is based on Hyperledger Fabric technology and methods at the database level, such as hash calculation of data, data fingerprint extraction, access control, and fast data reading. It can achieve multi-level data protection and data tamper-proofing through the entire life cycle of procurement transaction data and can effectively control the trust crisis that may be caused by the asymmetry of information among subjects of the procurement transaction platform. Finally, the feasibility of the prototype system for the proposed strategy is verified by the experimental analysis of the protection performance of the system.

Abstract:A Cesium-based three-dimensional (3D) visualization platform for space target observation is designed and developed to overcome the shortcomings of existing space target observation software in 3D visualization, cross-platform capability, and ease of operation and ultimately to meet the needs of stations in space target tracking and measurement. The software development framework is outlined, and orbit prediction is performed by leveraging two-line elements (TLE) and the latest simplified general perturbation model 4 (SGP4). Spatial coordinate transformation is conducted to achieve platform functions of orbit calculation of space targets, real-time position dynamic display, and station visibility analysis, and the prediction results are validated through the satellite tool kit (STK). The application results show that the proposed visualization software, with easy operation, favorable cross-platform performance, and calculation results that meet the accuracy requirements, is a useful exploration of combining the visualization and practicability of software design and development.

Abstract:This study designs and develops a representation and search method and a relevant system for dynamic linked data networks. The method can help users obtain the association around an entity when there are complicated entities and associations, and the network can be dynamically expanded through guided interaction. When users know multiple entities which are suspected to be associated, the minimum connected graph algorithm based on distributed computing is used to search out the association network. Application cases show that the proposed method and system can achieve good results and user experiences, and the system based on the method has been applied in many engineering projects such as Smart City, Safe City and Metropolitan IoT.

Abstract:Considering the internal driving mechanism of behavior decision-making and state changes of multiple UAVs, a collaborative trajectory planning method based on decision-making knowledge learning is proposed from the perspective of information processing. Firstly, the behavior states of UAVs are represented by knowledge on the basis of the Markov decision process, and the decision-making knowledge on continuous action space is developed. Then, a deep deterministic policy gradient (DDPG) algorithm based on decision-making knowledge learning is presented to achieve the collaborative planning of UAVs on the decision-making knowledge level. The experimental results reveal that on the basis of developing a demonstration system, the method can obtain an optimal trajectory planning strategy by reinforcement learning and can simultaneously achieve the convergence and stability of the comprehensive evaluation and average reward of trajectories, which provides decision-making support for mission execution of UAVs.

Abstract:Effective and safe data interaction across networks of different security levels is difficult due to the lack of a standard meteorological information transmission mechanism. Considering the diverse service application requirements, this study draws on the “2+1” model structure to design the architecture of the trusted interaction of meteorological service data streams across physically isolated networks. This architecture is then deployed in the demilitarized zones (DMZ) of the meteorological intranet and other networks to achieve safe data transmission and sharing across regions. After the trusted interaction architecture is outlined, application research is conducted according to specific meteorological service requirements. Finally, system function, performance, and security tests are carried out, and the transmission bottleneck and bandwidth utilization of the trusted interaction architecture are analyzed. This research can guide the practice of applying a trusted interaction architecture to improve the transmission efficiency of data streams across heterogeneous networks.

Abstract:Knowledge graph technology is used increasingly widely in industries. Therefore, it is more important to study its application in the field of geological reports to realize accurate queries and visualization for library users. Taking geological reports as the research object, this study uses crawler technology to obtain mineral, geographic area, organization and other entity information in geological reports. With the help of related technologies of knowledge graph, the study designs geological entities and relations for the knowledge graph of geological reports. After named entity recognition, relation extraction and attribute extraction, 266787 geological report entities and 306686 relations are constructed. Knowledge graphs are stored in the Neo4j graph database to improve the performance of geological data query and facilitate the query of library users. This research can provide theoretical support for the application of the knowledge graph to geological reports.

Abstract:In the design and development of the current conveyor line system, problems such as cumbersome work, poor flexibility for modification, and complicated on-site installation and debugging are common in links including the structure design and layout program planning, on-site installation and debugging of the control system, and real-time conveyor path planning of large-scale systems. Considering this, this study designs and develops a Unity3D-based simulation system for modular conveyor line design. On the basis of the three-dimensional (3D) model of the modular conveyor Unity3D, the modular and rapid assembly construction of the design scheme is realized through Winform human-computer interaction. The soft PLC technology and distributed controller architecture are used to achieve off-line programming and quick on-site installation of the conveyor line control system. The optimal path of material transportation is planned by the genetic algorithm, and the simulation operation of the conveyor line system is finally completed. The result comprehensively verifies the effectiveness and correctness of the system design scheme, control programming, and path planning and lays a foundation for the subsequent rapid design and on-site installation of conveyor lines.

Abstract:Two-dimensional (2D) face recognition is greatly affected by illumination, occlusion, and attitude. To overcome these shortcomings, this study proposes a 3D face recognition algorithm with multi-modal fusion based on deep learning. Firstly, the convolutional autoencoder fuses the color image and the depth map, and the fused image is input to the network for pre-training. In addition, a new loss function cluster loss is designed for pre-training in combination with the Softmax loss, so as to obtain a highly accurate model. Then, transfer learning is employed to fine-tune the pre-trained model, and thus a lightweight neural network model is obtained. The processed original dataset is used as the test set, and the identification accuracy of the test reaches 96.37%. Experimental results verify that the proposed method makes up for some shortcomings of 2D face recognition, and it is less affected by illumination and occlusion. Compared with 3D face recognition using high-precision 3D face images, the proposed algorithm is faster and more robust.

Abstract:With the rapid development of virtual reality technology, somatosensory sensors such as Leap Motion appear and are widely used in human-computer interaction. This study proposes a Leap Motion gesture interaction method based on a deep neural network to resolve the problem that the Leap Motion somatosensory controller has a low recognition rate and a slow recognition speed at the edge of its recognition range. In addition to the defined interactive gestures, a three-dimensional interactive system is designed and applied to a virtual scene. Specifically, the system captures data with Leap Motion, uses the deep neural network to extract features from the acquired infrared images, and implements gesture classification and recognition. Then, the changes in the hand coordinates between two adjacent frames acquired by Leap Motion are utilized to determine dynamic gestures. Finally, the interaction function in the virtual scene is fulfilled by investigating the dynamic gestures. Experimental verification shows that the proposed gesture recognition method is superior to the built-in gesture recognition method of Leap Motion in both recognition speed and recognition accuracy. Moreover, it still maintains a high recognition rate at the edge of Leap Motion’s recognition range.

Abstract:To accurately classify Sina microblog comment information, this study proposes an improved genetic algorithm-improved particle swarm optimization-balanced support vector machine (GA-IPSO-BSVM) classification model to enhance the accuracy and convergence of classifying Sina microblog comment information. Firstly, to effectively improve the algorithm convergence speed and efficiently save computational resources, this model introduces the elimination mechanism of the GA in the early iteration to remove a large number of low-speed particles. Secondly, to avoid the algorithm being trapped in local optima and improve the topology of particle relations in PSO, this study utilizes a K-means clustering algorithm to perform cluster partition of particle swarms in the middle of the iteration. The particle swarms are iterated in the communities and excellent particles are selected in each community. Thirdly, all excellent particles in the communities are combined into an excellent particle swarm that is iterated to derive the global optimal solution in the late iteration. Fourthly, the hyperparameter optimization of BSVM is performed by combining GA with IPSO to enhance classification accuracy. Finally, the proposed GA-IPSO-BSVM model is used for verifying the classification and prediction of Sina microblog comment information. The experimental results demonstrate the superiority of the proposed classification model over other benchmark models applied to Sina microblog comment information classification in terms of accuracy improvement.

Abstract:Single-object pedestrian tracking is one of the most basic and widely studied tasks in computer vision object tracking. However, most of the correlation filtering algorithms and deep learning algorithms currently used have insufficient tracking accuracy and real-time tracking performance. To solve the above problems, we propose a real-time single-object pedestrian tracking algorithm based on deep and shallow feature fusion. Firstly, this algorithm predicts the object location by Kalman filters and extracts the shallow color features of the object by calculating the four-part color histogram, and the prediction similarity is obtained to judge the reliability of prediction results. Then, YOLOv4 is used as a detector to extract deep features of the object and then calculate the distance metric of motion information and appearance information. Meanwhile, the shallow color features of the detection object are extracted to calculate the similarity distance metric, and the weighted fusion of the feature distance metric is employed to match the detection object and update the tracking trajectory. Finally, a trajectory updating strategy is put forward to coordinate the calling relationship between the prediction block and the detection block and to achieve a balance between tracking accuracy and speed. Testing experiments are conducted on the OTB100 and LaSOT datasets. The experimental results demonstrate that the tracking accuracy of the proposed algorithm on the above datasets reaches 0.581 and 0.453, respectively, and the tracking speed tested on GPU can achieve 33.64 FPS and 35.32 FPS, respectively, which meets the requirements of real-time tracking.

Abstract:The high-performance computing environment is designed to provide high-performance computing services for users and research teams. As more and more supercomputing centers, application communities, and service platforms have access to the environment, their users expect to use resources by logging in the high-performance computing environment with their original accounts. Now, only grid accounts authenticated by LDAP can access the application program interface (API) of the high-performance computing environment. To meet the expectation of users, we develop a new version of API for the high-performance computing environment. This study mainly introduces the structure and implementation of the new API, and the calling ways of the new API are exemplified. The new API can provide more convenient and secure services for the access of communities and service platforms to a high-performance computing environment.

Abstract:A real-time obstacle avoidance method for industrial vehicles based on binocular positioning and ranging is proposed to solve the problems of environmental influence and signal interference faced by the current obstacle avoidance technology of industrial vehicles. Firstly, the binocular depth camera is calibrated, and binocular stereo correction is performed on the images of the operating environment directly behind the vehicle. Secondly, the SGBM algorithm is used to calculate the parallax map, and the 3D point cloud reconstruction is carried out by the trigonometric transformation principle in combination with internal parameters of the camera. Next, the ground calibration and ground equation fitting are conducted, and the effective detection range and safety warning range are defined. Finally, the orientation of pedestrians is detected, and the distance calculation of pedestrians detected in the range is carried out by the algorithm of straight and turn ranging, and the range warning and obstacle avoidance are realized in real time. Four groups of experiments show that the errors of the pedestrian ranging algorithm are lower than 0.1 m and 0.2 m in 0–3 m and 3–5 m in straight and turning states, respectively. The identification accuracy of the pedestrian detection algorithm is 97.38%, and the detection frame rate is 22.12 fps. The method has high sensitivity within the set range and good real-time obstacle avoidance effects.

Abstract:The delayed access to indirect memory often affects the execution performance of applications. An effective solution is to resort to the prefetching technology. Although the Shenwei platform developed in China supports the software and hardware prefetching mechanisms for conventional access modes, the compilers in its GNU compiler collection (GCC) lack the method of automatically inserting prefetches for indirect memory access. A complete indirect prefetching optimization pass is developed on the basis of the Shenwei GCC to solve this problem, and it uses a depth-first search algorithm to find indirect memory references that refer to loop induction variables and generate appropriate software prefetches for them. In a set of memory-bound benchmark tests, the average speed-up ratio of the automatic prefetching pass on the SW1621 processor reaches 1.16 times.

Abstract:Intravoxel incoherent motion (IVIM) magnetic resonance imaging is a non-invasive technique, which can characterize the diffusion and perfusion of water molecules in biological tissues. Traditional IVIM parameters estimation methods are highly affected by the noise, and the parameter estimation is not effective. In order to accurately and quickly determine the diffusion and perfusion parameters in tissue regions, this study proposesd a one-dimensional dynamic convolutional neural network (DCNN) based on the dynamic convolutional module to estimate IVIM parameters. It takes into account the contextual information between the voxel signals and the contribution of b-values, to estimate IVIM parameters. The DCNN is compared with the traditional estimation method on the test simulation data and real acquisition images underwith different noise levels. The experimental results show that the proposed DCNN method can reduce the coefficient of variation, bias, and relative root mean square error of the IVIM parameters and, improve the parameter consistency and robustness, and have good visual quality at the same time.

Abstract:For color distortion and incomplete dehazing caused by inaccurate media transmittance in the image dehazing algorithm, an image dehazing algorithm with an improved residual neural network is proposed. First, a parallel multi-scale convolutional layer is adopted to extract the characteristics of the haze image. Then the media transmittance is learned by introducing the residual network of the depthwise separable convolutional layer and refined by the weighted guided filter. Finally, according to the atmospheric scattering model, a clear image without hazy is obtained. Experimental results show that compared with other dehazing algorithms, the proposed algorithm improves peak signal to noise ratio (PSNR) and structural similarity (SSIM) indicators, and the dehazing image also performs well in subjective vision.

Abstract:The release and analysis of multidimensional data can produce great value. However, privacy disclosure often occurs in the data collection phase. The traditional centralized differential privacy protection method requires a completely trusted third-party data collector, which is quite difficult to be found in practice. With the increase in attribute dimensions, the refinement of data collectors (the calculation of joint distribution) has also become an urgent problem to be solved. To address the above problems, this study proposes a localized differential privacy protection algorithm (RR-LDP) for multi-valued data. Unary coding and instantaneous random response technique are introduced to protect personal privacy in the data collection phase, which reduce communication overhead. With the combination of expectation maximization (EM) algorithm and LASSO regression model, the study puts forward an efficient joint distribution estimation algorithm (LREMH) for multidimensional data, which meets the requirement of LDP. The algorithm uses the LASSO regression model to estimate the initial value and employs the EM algorithm for iterative calculation. Theoretical analysis and experimental results show that the LREMH algorithm achieves a balance between accuracy and efficiency.

Abstract:The existing algorithms of circle clipping against an arbitrary polygon window suffers from complex steps and do not consider the case where the polygon contains an inner ring. Thus, this study presents a new algorithm based on the parameter analysis of intersection points for circle clipping against an arbitrary polygon window. In this algorithm, only by comparing the parameter values of the intersection points on the directional line of the edge, one can classify the intersection points into entry points and exit points. After the intersection points are sorted out, the arcs within the clipping window can be obtained with the combinations of “entry point $\Rightarrow $ exit point.” The proposed algorithm is proved feasible by programming results and is universal to circle clipping against an arbitrary polygon window, even if the polygon contains an inner ring.

Abstract:YOLOx-Darknet53 is an improved detection network integrating a basis of you only look once version 3 (YOLOv3) with various tricks added. Nevertheless, it still uses Darknet53 as the backbone network to extract features, so the feature extraction capability of the network is still insufficient. In this study, we acquire a contextual attention (CoA) module by improving the attention mechanism in CoTNet and replace the 3×3 convolution in the residual block of the YOLOx backbone network with the module to obtain a new residual block after attention fusion and thereby strengthen the feature extraction capability of the backbone network. A comparison experiment is conducted on the Pascal VOC2007 data set. The mean average precision AP@[.5:.95] and the AP@0.5 of the network integrating the CoA module are both 1.4 higher than those of the original network. After the backbone network is improved, a non-parameter 3D attention module is added in front of the YOLOx detection head to obtain the final improved detection network. The results of another round of the above comparative experiment show that the AP@[.5:.95] and the AP@0.5 of the final network are respectively 1.6 and 1.5 higher than those of the original network. Therefore, the improved network is more accurate than the original network in detection and can achieve better detection effects in industrial applications.

Abstract:Respiratory rate is one of the important indicators of human health. To solve the problems of the existing respiratory rate detection methods including one single human posture and poor detection accuracy and robustness, this study proposes a visual detection method of human respiratory rate suitable for multiple postures. This method uses an ordinary camera to capture human breathing videos. The image pyramid optical flow is used to process continuous video images and thereby obtain the moving foreground region, wherein the largest connected area is preliminarily identified as the thoracoabdominal breathing area. Then, the breathing region in each frame of the video is input into the complex steerable pyramid for multi-scale and multi-directional spatial decomposition, and amplitude spectra and phase spectra on multiple scales and in multiple directions are obtained. On this basis, the phase spectra on multiple scales and in multiple directions of each frame are weighted by the amplitude spectra and then averaged to obtain the phase-time signal. Finally, decisions are made for the extracted signal. If the dominant frequency of the signal is within the frequency band of the respiratory signal and the energy proportion is high, the respiratory rate is obtained by peak detection of the signal. Otherwise, continuous video images are reselected for subsequent detection. The experimental results show that this method is suitable for respiratory rate detection in various postures and that it is superior to the existing methods in accuracy and robustness.

Abstract:Image segmentation is the basis of computer-aided film reading, and the accuracy of wound image segmentation directly affects the results of wound analysis. However, the traditional method of wound image segmentation has cumbersome steps and low accuracy. At present, a few studies have applied deep learning to wound image segmentation, but they are all based on small data sets and can hardly give full play to the advantages of deep neural networks and further improve accuracy. Maximizing the advantages of deep learning in the field of image segmentation requires large data sets, but there is no large public data set on wound images as establishing large wound image data sets requires manual labeling, which consumes a lot of time and energy. In this study, a wound image segmentation method based on transfer learning is proposed. Specifically, the ResNet50 network is trained with a large public data set as a feature extractor, and then the feature extractor is connected with two parallel attention mechanisms for retraining with a small wound image data set. Experiments show that the segmentation results of this method are greatly improved in the average intersection over union (IoU), and this method solves the problem of low accuracy in wound image segmentation due to the lack of large wound image data sets to some extent.

Abstract:Flight simulators are important equipment for simulating and reproducing real flight activities, and the simulation effects of simulators have been attracting wide attention. However, the motion platform based on the classical washout algorithm for restoring motion trajectories faces problems such as conservative parameter settings and poor simulation effects. Therefore, this study proposes a filter parameter optimization method based on an improved artificial fish swarm algorithm. Specifically, by the human vestibular perception error model, the corresponding objective function is obtained; then, the improved fish swarm algorithm is used to optimize the natural cut-off frequency in the filter; finally, the optimized filter parameters are simulated and verified through the simulation model built on Simulink. The results show that compared with those of the classical washout algorithm and the basic artificial fish swarm algorithm, the new parameters obtained by the improved algorithm can effectively improve the motion perception effect during the algorithm washout, reduce the motion error, and save more motion space.

Abstract:To address the blindness and low efficiency in the inversion of air pollution sources, this study proposes a novel method for air pollution source inversion based on a modified ant colony optimization (M-ACO) algorithm. The Gaussian diffusion model for point source pollution is used to establish the pollution source inversion model which is solved by an ant colony optimization (ACO) algorithm. In view of the shortcomings in the ACO algorithm, the idea of selection and crossover in genetic algorithms is introduced to enrich population diversity, which thus avoids falling into local extrema. At the same time, a reward and punishment factor mechanism is designed to improve the pheromone update rule so that the algorithm can converge faster. It is then summarized as the M-ACO algorithm. Comparative experiments prove that the M-ACO algorithm can make the inversion results of pollution sources more accurate and efficient than the ACO algorithm. It provides effective theoretical support for the practical application of air pollution source inversion.

Abstract:Given the problems of color distortion and background noise amplification using contrast limited adaptive histogram equalization (CLAHE) to enhance X-ray images of airport security inspection, this study proposes an X-ray image enhancement algorithm based on three-level image fusion and CLAHE. Specifically, the X-ray image is converted into RGB and HSV images for CLAHE enhancement respectively. The enhanced images are fused by the Euclidean norm for the first-level fusion. Then, the fused images are sharpened by unsharp masking (USM), during which the second-level image fusion is performed according to the mask. Finally, the sharpened images and the original image are combined according to the coefficient to complete the third-level fusion. The simulation experiment results show that the proposed algorithm effectively improves the contrast of security inspection X-ray images, increases the average peak signal-to-noise ratio (PSNR) by 7 dB, and suppresses the color distortion and background noise in the enhanced images. This algorithm helps to improve the accuracy of identifying prohibited items in X-ray images and thus has a positive impact on the construction of a safe airport.

Abstract:Due to the increasing demand for data transmission anytime and anywhere, the low bandwidth and high jitter of the communication of a single heterogeneous wireless network seriously affect users’ experience of mobile devices. Considering the actual demand of mobile devices for link aggregation of heterogeneous wireless networks, especially the demand of Android mobile devices, this study proposes a scheme to realize link aggregation of mobile data networks and WiFi networks, two heterogeneous wireless networks, on mobile devices. The scheme is based on the application layer development, which skips the complex packet encapsulation process of the lower layer and does not change the hardware configuration of a mobile device. In this study, the working principle and design idea of wireless link aggregation are described, and the program is implemented and tested on Android phones. The test results show that the aggregation scheme of heterogeneous wireless links can significantly improve network bandwidth and file transfer speed.

Abstract:Although dictionary learning mostly uses linear functions to capture potential features of data, this method cannot fully extract the inherent feature structure of data. Deep learning has received widespread attention in recent years due to its outstanding feature representation ability. Therefore, this study proposes a nonlinear feature representation strategy combining deep learning with dictionary learning, i.e., deep neural network-based dictionary learning (DNNDL). DNNDL integrates the dictionary learning module into the traditional deep learning network structure and simultaneously learns the data dictionary and the sparse representation coefficients on it in the low-dimensional embedded space mapped by the autoencoder, thereby achieving end-to-end potential data feature extraction. It can generate compact and discriminant representations of existing data as well as out-of-sample point data. DNNDL not only is a new deep learning network structure but also can be regarded as a unified framework of dictionary learning and deep learning. A large number of experiments on four real data sets show that the proposed method has a better data representation capability than those of conventional methods.

Abstract:When no obvious boundaries exist between skin regions and non-skin regions, skin detection becomes extremely difficult. To solve this problem, we propose a new skin detection and correction algorithm. Firstly, this study uses a convolutional neural network (CNN) to extract skin features such as colors and texture step by step and then subdivides the boundary region of skin and non-skin pixels through the gated convolutional layer to enhance the effect of skin detection. Finally, ASPP is applied to fuse deep information and edge information. The detection results from rough threshold segmentation are used as input for the evaluation on ECU and Pratheepan datasets. The experimental results show that the accuracy of this algorithm reaches up to 91% on the ECU dataset and 95% on the Pratheepan dataset. The performance of the proposed algorithm has been significantly improved compared with that of the existing methods.

Abstract:In recent years, obtaining multi-modality magnetic resonance (MR) images with automatic generation methods has been widely studied. However, it is still difficult to generate images of all the other modalities by one given modality. To solve this problem, this study proposes a dynamic generative adversarial network (DyGAN) model. By combining the generative adversarial network and dynamic convolution and introducing a task label, the new model can simultaneously generate other three MR modalities from one modality. In addition, a multi-scale discrimination strategy is further proposed to improve the quality of image generation by fusing multiple scales. Image generation is verified on the BRATS19 dataset. The experimental results show that the new method can not only simultaneously generate multi-modality images but also improve the quality of the generated images.

Abstract:In view of the sparse user check-in data, underutilization of review text information, and low accuracy of point-of-interest recommendation in location-based social networks, this study proposes a point-of-interest recommendation model based on review texts and the convolutional neural network (CNN), or an RT-CNN point-of-interest recommendation model for short. To start with, the Gaussian function and adjacent geographical location weighting are used to fill in the missing location information in the matrix decomposition model and thereby predict the user’s potential interest in unchecked locations. Then, review text information is processed by the convolutional neural network to mine potential features and ultimately to extract the user’s emotional tendencies in depth. The Softmax logic regression function is utilized to obtain the probabilities of the review text related to the potential features of a user and a point-of-interest location, and potential feature vectors of the user and the location are extracted by solving the objective function. Finally, the check-in behavior, geographical location influence, user emotion tendencies, user potential features, and potential features of point-of-interest locations are integrated to recommend points-of-interest. Experiments are carried out on two real check-in datasets, namely NYC and LA, on the public website Foursquare. The results show that compared with other state-of-the-art point-of-interest recommendation models, the RT-CNN model improves the accuracy rate and the recall rate and has better recommendation performance.

Abstract:Compared with other fine-grained image classifications, that of wild snakes is more difficult and complicated, as it is difficult to judge and classify snakes by their local characteristics due to their different postures, rapid posture changes, and usual status of motion or coiling. In response, this study applies the self-attention mechanism to fine-grained wild snake image classification to solve the problem that the convolutional neural network focuses too much on the local parts to ignore the global information due to the increasing number of layers. Transfer learning is implemented through Swin Transformer (Swin-T) to obtain a fine-grained feature extraction model. To further study the performance of the self-attention mechanism in meta-learning, this study improves the feature extraction model, builds a Siamese network, and construct a meta-learner to learn and classify a small number of samples. Compared with other methods, the proposed method reduces the time and space consumption caused by feature extraction, improves the accuracy and efficiency of meta-learning classification, and increases the learning autonomy of meta-learning.

Abstract:The collaboration network of mobile edge computing (MEC) and device-to-device (D2D) technology takes into consideration multiple devices, where the final output of multiple wireless devices is used as the input of a subtask on another device. The optimal resource allocation (offloading transmit power and local CPU frequency) and task offloading decisions are studied to minimize the weighted sum of the energy consumption of wireless devices and the task completion time. First, given an offloading decision, the closed expression of offloading transmit power and local CPU frequency are derived, and the convex optimization method is used to find the solution to the problem. Then, on the basis of the one-climb policy, a low-complexity linear search algorithm is proposed, which can obtain the best offloading decision in linear time. Numerical results show that the performance of this strategy is significantly better than that of other representative benchmark tests.

Abstract:Electronic medical records are the archives to note patients’ health conditions during treatment, where a large number of medical entities are scattered throughout the text and a wealth of medical information is contained. Existing relation extraction models in the medical field mainly utilize the relation classification method to recognize the semantic relation between two medical entities. Chinese electronic medical records have the characteristic of a dense distribution of medical entities in the text. In response, this study proposes a method based on condition hint and sequence labeling to extract relation triples. In this approach, the relation triple recognition task is converted to a sequence labeling task. The head entity and relation type in a relation triple combine to form condition hint information, and the model recognizes tail entities relevant to the condition hint information from the text of electronic medical records by sequence labeling. The experimental results on an electronic medical record dataset show that this method can be applied to recognize relation triples in Chinese electronic medical records.

Abstract:Bearing fault diagnosis plays a vital role in maintaining rotating machinery and avoiding major disasters. Given that the existing fault diagnosis model cannot adapt to the changing working loads in actual industrial applications, a fault diagnosis method based on feature fusion and hybrid enhancement is proposed. For this purpose, new feature signals are generated by fusing time-frequency features, working condition features, and time difference features into the original signal. Then, the phase space reconstruction theory is applied to convert the feature signals into image signals, and data distribution is expanded through hybrid enhancement during training. Finally, the residual network is used for fault diagnosis analysis. The experimental results on the Case Western Reserve University (CWRU) dataset show that the prediction accuracy of this method under invariable working conditions is up to 100% and its average prediction accuracy under changing working conditions reaches 93.28%, which indicates that the proposed method has a remarkable domain adaptability.

Abstract:Although the separation of the devices in the control layer and the forwarding layer can be achieved by software-defined networking (SDN), the decoupling of the two layers exposes the devices in different layers of the network to new types of distributed denial of service (DDoS) attacks. To solve the above problem, this study proposes a DDoS attack detection method based on the improved Dempster-Shafer (D-S) theory for detecting DDoS attacks aimed at SDN controllers and switches in an SDN environment. In the improved algorithm, the discrete factor and the purity factor are used to measure the conflicts among D-S evidence sources. Meanwhile, the evidence sources of the D-S evidence theory are adjusted according to the two factors, and the DDoS attack detection result is obtained with the adjusted evidence sources in light of Dempster’s rule of combination. Experimental results show that the proposed method achieves high detection precision.

Abstract:To address the cold-start and sparsity problems of recommendation systems, this study proposes a recommendation model based on a heterogeneous information network. Previous approaches are unable to take into account both knowledge graph representation learning and implicit path information, which makes the performance of knowledge recommendation systems mediocre. The proposed method sets meta-paths in the heterogeneous information network and integrates them into knowledge graph representation learning by the graph neural network (GNN). Next, the attention network is used to connect a recommendation task with a knowledge graph representation task. It can not only learn the potential features of the two tasks but also enhance the interactions between the recommended items in the recommendation system and the entities in the knowledge graph. Finally, the user click rate is predicted in the recommendation task. The method is experimented on the open dataset Book-Crossing and the knowledge graph constructed with the DBLP dataset, and the results demonstrate that the proposed model achieves better performance than that of other algorithms in indexes of area under curve (AUC), recall, and F1-score.

Abstract:Electroencephalogram (EEG) generation via generative adversarial networks (GANs) suffers from various issues including invariant features of samples generated, large amplitude differences, and slow fitting speeds. The quality of signals thus generated fails to meet the requirements of deep-learning model training and optimization. To address the issues above, this study optimizes the Wasserstein GAN gradient penalty (WGAN-GP) so that it can perform better in EEG generation. The details are as follows: (1) On the basis of the framework of the WGAN-GP network, the convolutional neural network (CNN) is replaced by the long short-term memory (LSTM) network to ensure the integrity of time-dependent features and thereby solve the issue of invariant features; (2) real EEGs are normalized and then applied to the discriminator to reduce the amplitude differences; (3) the noisy parts of EEGs are applied to the generator as prior knowledge to increase the fitting speed of the generation model. Sliced Wasserstein distance (SWD), mode score (MS), and EEGNet are applied to evaluate the proposed generation model quantitatively and hierarchically. Compared with the current generative network WGAN-GP, the proposed model provides data closer to their real counterparts.

Abstract:The coronavirus herd immunity optimization (CHIO) algorithm is improved to form a hybrid algorithm for the permutation flow-shop scheduling problem (PFSP). Specifically, in the stage of herd immunity evolution, the strategy of dynamically changing the expansion rate is used to balance the exploration and developemnt ability of the algorithm. After the rebirth stage, a crossover stage based on differential evolution is added to enhance the mining ability of optimal solutions. The solution to PFSP is encoded and decoded by the smallest position value to minimize the maximum completion time. The experiments on 21 Reeves test examples indicate that the proposed algorithm is effective in solving PFSP.

Abstract:The driver assistance system is considered the first choice for solving traffic safety problems. The basis of developing a driver assistance system is to accurately recognize the vehicle behavior for applications in aspects such as vehicle safety warning, path planning, and intelligent navigation. The existing behavior recognition methods based on the support vector machine model, hidden Markov model, and convolutional neural network still face the imbalance problem between calculation amount and accuracy. This study proposes a Gaussian mixture hidden Markov model, which is a combination of the hidden Markov model and the Gaussian mixture model. The model is experimentally verified on the NGSIM data set from the Federal Highway Administration of the USA, and the results reveal that the model has higher accuracy in the recognition of free lane-changing behavior. Additionally, this study optimizes the parameters of the proposed model to achieve the best recognition effect and provide a reference for the vehicle behavior recognition of intelligent driving in the future.

Abstract:The rapid development of smartphones and smart operating systems boosts the prevalence of natural language conversations in human-machine interactions. In the case of multiple-function tasks, however, the conversation system will generate a complex task command, and a variety of problems will arise. The current NLP technology can provide some solutions, but its capability to dynamically recognize and process task commands is insufficient in solving complex problems. In this study, we propose a solution that combines the NLP engine and task scheduling unit. Specifically, natural language commands are used for task scheduling, and thus the conversation system can accurately recognize command tasks and related parameters and generate a rational schedule for the tasks. In addition, a conversation strategy is proposed to address ambiguity or information omission in the natural language conversation, by which conversation information can be collected with minimum question-answering iterations when necessary.

Abstract:Given that the conventional method of weighing glass gobs has a low measurement efficiency and is greatly influenced by the environment, this study proposes a non-contact measurement method based on binocular vision. For this purpose, a binocular vision system is built to filter and de-noise the collected images and extract their feature contours. A disparity map with complete edge information is obtained with the Census transform-based stereo matching algorithm integrating the gradient information of the gob images. The influences of gobs deflecting along the directions of the camera plane and the camera depth of focus, respectively, on the precision of horizontal slicing are analyzed. Specifically, the minimum bounding rectangle algorithm is utilized to correct gobs deflecting along the direction of the camera plane. Then, disparity information is used to correct gobs deflecting along the other direction. Finally, horizontal slices are accumulated through horizontal slicing to obtain the gob volume and mass. The experimental results show that the proposed method also reaches the precision standard on gobs with spatial deflection and it thus meets the demand of glass bottle production.