Proceedings Volume 12076

2021 International Conference on Image, Video Processing, and Artificial Intelligence

cover
Proceedings Volume 12076

2021 International Conference on Image, Video Processing, and Artificial Intelligence

Purchase the printed version of this volume at proceedings.com or access the digital version at SPIE Digital Library.

Volume Details

Date Published: 11 November 2021
Contents: 4 Sessions, 43 Papers, 0 Presentations
Conference: Fourth International Conference on Image, Video Processing, and Artificial Intelligence (IVPAI 2021) 2021
Volume Number: 12076

Table of Contents

icon_mobile_dropdown

Table of Contents

All links to SPIE Proceedings will open in the SPIE Digital Library. external link icon
View Session icon_mobile_dropdown
  • Front Matter: Volume 12076
  • Computer Vision and Image, Speech, and Signal Processing
  • Pattern Recognition and Machine Learning
  • Data Science and Artificial Intelligence Applications
Front Matter: Volume 12076
icon_mobile_dropdown
Front Matter: Volume 12076
This PDF file contains the front matter associated with SPIE Proceedings Volume 12076 including the Title Page, Copyright information, and Table of Contents.
Computer Vision and Image, Speech, and Signal Processing
icon_mobile_dropdown
Learning human-object interaction detection via deformable transformer
The goal of human-object interaction (HOI) detection is to localize both the human and object in a picture and recognize the interactions between them. HOIs are always scattering in the image. The traditional methods based on CNNs are unable to aggregate the information scattered in the image. Many new methods utilizing the contextual features cropped from the outputs of the CNNs, which sometimes are not effective enough. To overcome the challenge, we utilize the deformable transformer to aggregate the whole features output form the CNNs. The attention mechanism and query-based predictions are the keys. In view of the success of the methods based on graph neural networks, the attention mechanism is proved to be effective to aggregate the contextual information image-wide. The queries can extract the features of each human-object pair without mixing up the features of other instances. The deformable transformer can extract effective embeddings and the prediction heads can be fairly simple. Experimental results show that the proposed method is effective in HOI detection.
UD-GAN: a dense connected generative adversarial network for pixel-level modal translation of multimodal images
Modal translation between multimodal images is an effective complementary scheme when images with some certain modal are difficult to obtain. Since pixel level image modal translation method can obtain the images with the highest quality, it has become a research hotspot in recent years. Generative adversarial network (GAN) is a network for image generation, due to the complex structure of GAN and the complexity of the image generation task, the training results of GAN are not stable. In this paper, on the basis of U-nets, the dense block is used to increase the feature information in the subsampling encoding and up-sampling decoding operation, so as to reduce the loss of information and obtain higher quality images. At the same time, the dense long connection is introduced to connect the encoding and decoding operations of the same stage, so that the network can effectively combine the features at low and high level, and improve the performance of the network. Experimental results show that the proposed method is effective in modal translation of multimodal images, and the image quality is better than some state-of-the-art methods.
An end-to-end multi-task learning to link framework for emotion-cause pair extraction
Haolin Song, Dawei Song
Emotion-cause pair extraction (ECPE), as an emergent natural language processing task, aims at jointly investigating emotions and their underlying causes in documents. It extends the previous emotion cause extraction (ECE) task, yet without requiring a set of pre-given emotion clauses as in ECE. To solve ECPE task, we regards emotion-cause pair extraction as a link prediction task, and learns to link from emotion clauses to cause clauses, i.e., the links are directional. We propose a multi-task learning model that can extract emotions, causes and emotion-cause pairs simultaneously in an end-to-end manner. Emotion extraction and cause extraction are incorporated into the model as auxiliary tasks, which further boost the pair extraction. Experiments are conducted on an ECPE benchmarking dataset. The results show that our proposed model outperforms a range of state-of-the-art approaches.
EEG unilateral limb motor imagery modeling based on fMRI screening
Jun Ma, Wen Wang, Wenzheng Qiu, et al.
Motor imagery brain computer interface (MI-BCI) recognizes brain motor intention through electroencephalogram (EEG) acquisition and deep learning. The advantage of MI-BCI is that the recognition of brain ideas does not depend on task prompts, but it is difficult to accurately recognize the unilateral limb motor imagery tasks because of the difficulty of EEG decoding algorithm. In this paper, the asynchronous functional magnetic resonance imaging (fMRI) and EEG motor imagery data of unilateral limb hand grasping and hand handling tasks are creatively collected, and the brain activation features of each task are obtained by fMRI statistical analysis. The activation difference of the fMRI cerebral cortex is mapped to the corresponding channel position of the corresponding EEG and compared with the average power spectrum density (PSD) of each channel of each EEG trail. According to the comparison results, the more consistent EEG data are selected for training. The experimental results show that the screened EEG data training model shows a predict accuracy of 69.3%, which is a better classification result. The results proved that screening high-quality EEG by fMRI data has a certain effect, and the model is more consistent with the characteristics of brain motor imagery. This method can improve the prediction accuracy and guide the subjects to imagine correctly.
Ultrasonic phased array sparse-TFM imaging based on deep learning and genetic algorithm
Junying Song, Yanyan Liu, Shiwei Ma
As ultrasonic phased array total focus method (TFM) imaging technology can achieve full range of dynamic focusing with clear imaging and strong ability to characterize defects, TFM imaging algorithm has become the gold standard for testing other post-processing algorithms. However, due to the large amount of data and time-consuming calculation, the TFM imaging has limited the application in some industrial fields. With the reduction of the number of phased array transmitting elements, the imaging quality gets worse. To improve the imaging efficiency of the TFM algorithm and ensure the imaging quality, this paper proposes a method combining Siamese Convolutional Neural Network (SCNN) and Genetic Algorithm (GA) to obtain an optimal sparse array layout, to approximate the imaging effect of the full array with limited effective elements. After selection of an appropriate sparsity ratio, GA is used to optimize the sparse array layout. The sparse array elements emit ultrasonic waves, and the full array elements receive echo signals for imaging. SCNN is trained by a self-built industrial defect dataset, to output the similarity between the sparse-TFM image and the full-array TFM image. The similarity is used as an objective evaluation index to evaluate the imaging effect. The optimal sparse array layout is proposed by combining subjective evaluation with objective evaluation.
Low-illumination environment vehicle detection based on optimized adaptive image enhancement
Kun Zhao, Jian Liu, Xiaocheng Xiao
With the increasing number of vehicles, vehicle detection has become an important part of intelligent transportation system. At present, most detection algorithms are only suitable for normal light conditions, but the detection performance is poor for low illumination conditions. In order to achieve effective vehicle detection under low illumination conditions, this paper proposes an image enhancement algorithm to increase the contrast of the image, thereby greatly improving the effect of vehicle detection. First, the image contrast is enhanced through an adaptive contrast stretching algorithm. Secondly, the bilateral filtering algorithm is used to filter out the noise in the image. Finally, the detection system based on Haar features and AdaBoost classifier is used to detect vehicles. Experimental results show that the proposed algorithm can effectively enhance image contrast, highlight vehicle information, and the vehicle detection accuracy rate under low illumination conditions reaches 87.04%.
A defect detection method for glass bottle mouth based on area segmentation
Junxu Lu, Xu Zhang
Quality inspection is an essential technology in the glass product industry. Machine vision has shown more significant potential than manual inspection at present. However, the visual inspection of the bottle for defects remains a challenging task in a quality-controlled due to the difficulty in detecting screw thread defects. To overcome the problem, we propose a bottle mouth detection method based on area segmentation. First, an area segment method with traditional image processing methods, which is based on the characteristics of screw thread, was proposed. According to the result of the segment, the bottle area is divided into a screw thread area and a non-screw thread area. For the former, a defect detection method that uses edge detection and Gaussian filters is proposed to precisely detect screw thread defects. For the latter, a defect detection method that uses techniques such as the Sobel algorithm and global binarization is proposed to precisely detect other defects. The proposed method is tested for data sets obtained by our designed vision system. The experimental result demonstrates that our framework achieves good performance.
Detection method of power line under uneven lightness for flying-walking power line inspection robot
Bo Li Sr., Xinyan Qin, Jin Lei, et al.
Due to the variation of sunlight conditions resulting in uneven lightness in images, the details of target objects tend to hide in the dark or bright regions, which is adverse to following image processing. To reliably land on the power line under the change of lightness, the flying-walking power line inspection robot (FPLIR) needs reliable detection for the power line. In this paper, a machine vision-based detection method of power line is proposed to adapt different lightness. Firstly, a visual system of the FPLIR is designed to collect and process power line images. Secondly, the multi-scale retinex (MSR) algorithm is used to reduce the influence of lightness. Then, the local binary pattern (LBP) map of power line image is generated by the LBP operator and is divided into many blocks. An LBP histogram vector is calculated for every block, then the first-order entropy and second-order entropy of every histogram vector are calculated. Finally, the first-order entropy, the second-order entropy, and the edge density of power line image are used as the feature vector of fuzzy c-means (FCM) to obtain the power line region. The experimental result shows that the accuracy of the proposed method is 82.6%, which is 9.3% more than the method without image enhancement. Thus, the proposed method can effectively detect power line, improving the robustness and accuracy of power line detection (PLD) during the FPLIR landing.
The novel G-CPDA detector for the MIMO-OCDM system
Yufang Yin, Xinyi Yang
The orthogonal chirp division multiplexing (OCDM), proposed in 2016, is a relatively new multi-carrier technology which has been studied in recent years. The complex probabilistic data association (CPDA) algorithm has been applied in the MIMO OCDM system which obtains better bit error rate (BER) performance than the minimal mean square error (MMSE) algorithm. To reduce the computational cost, this paper presents a novel group-based CPDA (G-CPDA) scheme and presents the simulation results.
Transfer learning for micro-expression recognition based on the difference key frame images
Zhihua Xie, Le Wang, Ling Shi, et al.
Micro-expression, revealing the true emotions and motives, attracts extraordinary attention on automatic facial microexpression recognition (MER). The main challenge of MER is large-scale datasets unavailable to support deep learning training. To this end, this paper proposes an end-to-end transfer model for facial MER based on the difference images. Compared with micro-expression dataset, macro-expression dataset has more samples and is easy to train for deep neural network. Thus, we pre-train the resnet-18 network on relatively large expression datasets to get the good initial backbone module. Then, the difference images based on adaptive key frame is applied to get MER related feature representation for the module input. Finally, the preprocessing difference images are feed into the pre-trained resent-18 network for fine-tuning. Consequently, the proposed method achieves the recognition rates of 74.39% and 76.22% on the CASME2 and SMIC databases, respectively. The experimental results show that the difference image between the onset and key frame can improve the transfer training performance on resnet-18, the proposed MER method outperforms the methods based on traditional hand-crafted features and deep neural networks.
Slipping control of flying-walking power transmission line inspection robot using ANFIS
Zhaojun Li, Jin Lei, Xinyan Qin, et al.
Power line inspection robot can replace manual inspection, reduce the burden of manpower, and collect more information about the state of power line. When power line inspection robot walks along the power line, it is prone to slipping of driving wheel, which causes some problems, e.g., inaccurate position recognition, unstable image acquisition. To solve the slipping problem on power line, a walking mechanism of a flying-walking power line inspection robot (FPLIR) is designed in this paper, working principle of the walking mechanism is elaborated, and its mathematical model is established. To determine the walking state of the robot on the power line, a slipping identification method is proposed by defining a slipping degree to evaluate the slipping state. Considering that the factors causing slipping have a high degree of uncertainty and cannot be described by a very precise model, we design an adaptive neural fuzzy controller using ANFIS for the walking mechanism. A rigid-flexible coupling model is established using ADAMS software to simulate and verify the walking process of the FPLIR along the power line, and effect of the controller is tested based on the evaluation of the slipping degree. These simulation results show that the proposed adaptive neural fuzzy controller can effectively restrain slipping and improve the stability of the FPLIR inspection. The walking efficiency of the FPLIR is increased 35% using the proposed controller in the simulation. This study provides important reference of direction and technology for the subsequent slipping control research of inspection robot.
Superpixel segmentation with attention convolution neural network
Jingjing Wang, Zhenye Luan, Zishu Yu, et al.
A superpixel consists of a series of small regions composed of pixel points that are located next to each other and have similar features such as color, luminance, and texture. Most of these small areas retain the original information of the image, which facilitates faster follow-up processing. Most of the existing superpixel segmentation methods do not use deep learning network architectures. Some methods use deep learning, but they also have a very simple network architecture. And there are a few superpixel segmentation methods combined with deep learning. In this method, a more complex convolutional neural network with an attention module is applied to superpixel segmentation. The resulting superpixel segmentation network is more complex, and the addition of the attention module allows for more accurate chunking of the images, thus yielding more comprehensive and detailed segmentation results. By experimenting on a public dataset BSDS500, the method has higher accuracy in superpixel segmentation. Also, the segmentation speed of the method is similar to that of the existing simple segmentation networks.
DDGANA: dual-discriminator GAN with attention module for infrared and visible image fusion
Jinwen Ren, Jingjing Wang, Hongzhen Li, et al.
Infrared images can be differentiated between target and background according to the distinction in thermal radiation, which generally works under any lighting condition. In comparison, visible images available in abundant detail, consistent as human visual system. In this study, a novel approach for the fusion of infrared and visible images is proposed built on a dual-discriminator GAN with attention module (DDGANA). Our approach establishes confrontation training between one generator and two discriminators. The goal of the generator is to output result with contrast information and details. The two discriminants aim to increase the similarity between the images generated by the generator and the infrared and visible images. After sequential adversarial training, DDGANA outputs images that have preserved the high contrast and the abundant texture detail. Experiments on the TNO dataset prove that our approach obtains an improved performance over the other approaches.
DA-ResUNet: a novel method for brain tumor segmentation based on U-Net with residual block and CBAM
Jingjing Wang, Zishu Yu, Zhenye Luan, et al.
This paper focuses on a new method for brain tumor segmentation in Magnetic Resonance Imaging (MRI), using a modified residual block and CBAM for the U-Net network. To deepen the network, we replace the convolutional layer with a residual block with a CBAM module. We also insert the CBAM dual-attention module after skip connection and upsampling at each layer. It solves the problem that the low-level features contain a lot of redundant information because the skip connection connects the feature maps extracted by the encoder directly to the corresponding layer of the decoder. The performance is evaluated on the MRI dataset of Medical Image Computing and Computer Aided Intervention Society (MICCAI) 2018 Brain Tumor Segmentation Challenge. Numerical results are presented in the form of Specifity, Sensitivity, HD_95 and Dice coefficient (DICE) for GD-enhancing tumor (ET), tumor core (TC) and whole tumor (WT), respectively. We compare the proposed method with expert manual method and other state-of- art methods. Experiments show that RDAU-Net achieves state-of-the-art performance.
Computer simulations for the denoising effect of morphological reconstructions for CT images on hexagonal grids and regular hexagonal regions
Xiqiang Zheng
In this paper, morphological operations on hexagonal lattices and regular hexagonal regions are applied to 2D parallel beam computerized tomography (CT) image denoising. To show some denoising effects of the morphological operations, for an input image, we add certain amount of noise to get the noised image; and perform Radon transform and inverse Radon transform on the noised image to obtain a CT image. To compare hexagonal lattices with square lattices fairly, the CT image is resampled to an image defined on a certain randomized grid, which is in turn resampled to an image defined on a hexagonal lattice and an image defined on the corresponding square lattice (with the same sampling rate), respectively. For the morphological operations, we use a structuring element defined on a hexagonal lattice and a structuring element defined on a square lattice such that those two structuring elements have the same number of ones. Each of those two images performs an erosion with the corresponding structuring element. Then use those two images as masks and use the two corresponding eroded images as makers to perform morphological image reconstructions to obtain the two reconstructed images. Finally, both usual and Phantom images are used as test images. For many different tested noise levels, the output data show that the image reconstructed on the hexagonal lattice is usually more similar to the input image than the one reconstructed on the square lattice.
Informetric analysis of researches on application of artificial intelligence in COVID-19 prevention and control
Zhuozhu Liu, Sijing Chen, Qing Han
The COVID-19 (2019 novel Coronavirus) is the most widespread pandemic infectious disease encountered in human history. Its economic losses and the number of countries involved rank first in the history of human viruses. Since the outbreak of the COVID-19 pandemic around the world, AI has made a great contribution to the prevention and control of the COVID-19 pandemic. In this paper, researches on the application of artificial intelligence in COVID-19 pandemic prevention and control were analyzed by informetric method. 432 papers indexed in Thomson Reuters’s Web of Science were studied by the perspectives of categories of researches, high frequency keywords, authors, institutions, journals and countries, and we get conclusions as follows: The analysis of keywords cooccurence shows application of machine learning and deep learning in COVID-19 pandemic diagnosis and prediction. The journal that received the most cites was (Radiology) and the journal that published the most papers was (Journal of Medical Internet Research). USA, India and China have the largest number of published articles. USA, China and UK are most influential countries. We also analyzed the review literature on the application of AI in COVID-19 pandemic prevention and control in the Web of Science, and found that these papers specifically can be divided into the following three categories: The first is the application of AI in clinical diagnosis and treatment, the second is the application of AI in the development of anti-epidemic drugs, and the third is the role of AI in the epidemiological research of COVID-19 and the social governance of pandemic prevention and control.
Low light image enhancement method based on guided filtering
Bin Yao, Zhen Han, Shiying Kang, et al.
Low light image enhancement methods based on classic Retinex model attempts to estimate the illumination component and to project it back to the corresponding reflectance. Therefore, the accuracy of illumination component estimated determines the performance of enhancement results. Based on Retinex model, this paper proposed an illumination component estimating method by guided filtering. In the proposed method, bright channel is used for obtaining a rough illumination map. Then, the gray image converted by the input low light image is employed as the guidance image and we refine the rough illumination map for obtaining a structural-awared illumination map by guided filtering. Lastly, the enhanced image can be achieved by a simple computation. Experimental results demonstrated that the performance of the proposed method significantly overpasses conventional low light image enhancement methods.
Online metering model and experiment for gas-liquid phase in oil production well based on nozzle pressure difference
Yong Xiang, Xinming Zhao, Jianying Zhu, et al.
On-line measurement of multiphase flow holdup in oil well is important to production performance analysis and measure optimization. In this paper, aiming at the problem of large measurement error in the process of gas-liquid two-phase measurement, an on-line metering model of gas-liquid flow based on nozzle differential pressure is proposed, and experimental research on two-phase flow measurement is carried out. The relationship between differential pressure fluctuation and parameters of gas-liquid phase flow is explored by testing experiments. The characteristic value of differential pressure fluctuation signal is analyzed, and relationship models between some dimensionless parameters and gas rate are established. According to the experimental data, the correlation between the differential pressure and flow rate of each phase is obtained by multiple regression. Experimental study shows that measurement error of gas rate is less than 10% and error of liquid content is less than 8% based on the regression of experimental test samples, which improves the application scope and accuracy of wellhead production measurement, and provides a new solution for the measurement of multiphase flow in high gas liquid ratio wells.
Image inpainting method based sparse analysis model of synchronous dictionary learning
Bin Li, Baohua Sun, Dekun Li, et al.
A novel image inpainting algorithm based on sparse analysis model is proposed. The model is formulated on an analysis dictionary. The dictionary is updated using a least-squares method. The experiments on images demonstrate improved performance in peak signal to noise ratio (PSNR) compared to other image inpainting methods including AMLE Inpainting, Harmonic Inpainting, Mumford-Shah Inpainting, and Transport Inpainting algorithm. The evaluation results showed that our proposed algorithm has better performance.
Research on a personalized physiological status cluster discrimination system based on pulse signal
Xiaoping Jiang, Xiaofei Chen, Chunman Wang, et al.
At present, the workload of mental workers in society is getting heavier and heavier, and it is necessary to conduct research on their physiological state assessment. Considering that the pulse is easily available and non-invasive, the pulse wave signal is used as the data source. The infrared pulse sensor is used to collect the pulse wave signals of the human body in different states, and the data is preprocessed and feature extraction is performed, and then the feature value weight is calculated by the Relief algorithm to obtain a new weighted pulse sample. The third step is to use k-means algorithm and fuzzy C-means (FCM) algorithm for clustering analysis of the weighted eigenvalue samples, and use the three parameters of error square sum, contour coefficient, and CH coefficient as clustering effect evaluation indicators. Finally, The state of the clustered model is determined according to specific eigenvalues, and the state classification is realized. Thus, a personalized physiological state classification method is proposed, which is to establish different classification algorithm models for different individuals, thereby avoiding Individual differences.
Research on the method of detecting and eliminating gross errors in Beidou monitoring time series
Qing An, Lang Rao, Shusen Wu
In deformation monitoring work, there are often gross errors in the monitoring data sequence. Due to the complex distribution of the monitoring data sequence, the conventional gross error detection method is not ideal in the processing of the monitoring data sequence. Based on this, the improved 3σ method based on wavelet analysis is proposed to achieve the purpose of identifying and eliminating gross errors. The experimental results show that the method can identify and eliminate the gross error signals that are randomly added, and it can fit the overall trend of the detail signal and the data signal well. The simulation example of the algorithm illustrates its trend extraction. The accuracy fully meets the requirements of gross error identification and elimination.
Research and application of bridge safety monitoring system based on GNSS
Qing An, Lang Rao, Shusen Wu
Based on high-precision GNSS / GIS integration technology, a set of bridge health monitoring system is designed. Taking Wuhan Baishazhou Yangtze River Bridge as an application demonstration, the functions of real-time deformation monitoring, comprehensive analysis, early warning prediction and health assessment of the monitoring station are realized by receiving GNSS satellite positioning signal and differential signal of reference station for precise calculation. The results show that the monitoring system can realize the full-automatic, high-precision, all-weather, three-dimensional deformation monitoring which is difficult to realize by traditional monitoring methods, and the positioning accuracy reaches the real-time centimeter level and the post millimeter level. The application of the system significantly improves the accuracy of bridge health monitoring and the reliability of prediction and evaluation.
Pattern Recognition and Machine Learning
icon_mobile_dropdown
A systematic review of automated reconstruction of indoor scenes using Point Clouds
Creating structured 3D modeling of indoor environment from captured data plays a vital role in applications such as indoor navigation and interior remodeling. The recent surge in indoor scene reconstruction from point cloud information has made substantial progress when dealing with the complexity and variability of the indoor environment but many open research problems remain. In this survey, a review of the existing methodologies in characterization of the input sources, reconstruction model along with the other components in the structured reconstruction pipeline is done. An overview of the reconstruction methodologies is summarized along with its effectiveness and limitations while also acknowledging future research directions and challenges for the purpose of promoting research interest.
Query interface schema extracting from deep web using ontology
There are abundant information on the network, such as weather information, book information and so on. Now the Web information has far exceeded human imagination. However, most deep web information resources are not available through simple click, and often need query interface. We use domain ontology to construct the interface schema extraction method, the approach extracts interface schema from the perspective of visual units and the internal codes. There are many data extraction methods. Among them, CAW (Context-Aware Wrapper) is representative and outstanding. Thus, we choose it as our state-of-art baseline algorithm. The extraction effect of this method is better than CAW in the experiment. Finally, the effectiveness of this method is verified.
Adaptive energy-load matching optimization method
Dexin Li, Shang Wang, Chang Liu, et al.
Due to serious new energy abandonment in Jilin grid of China, an adaptive energy-load matching optimization method for wind-photovoltaic complementary is built in this paper. Among certain new energy sources, we adaptively select the combination of new energy stations with high load matching, so that more new energy is consumed and less power is discarded. In this paper, the optimization method of maximizing new energy consumption without energy-load matching is taken as the comparison benchmark, and the advantages of adaptive energy-load matching optimization method are analyzed from multiple scenarios of different total amount of new energy. By selecting the combination of new energy stations with high load matching degree, the method can obtain lower curtailment of new energy, and the complexity of decision space is reduced, making the decision variables of new energy stations can be solved in a short time,when the number of decision variables is big.
Atmospheric visibility prediction based on multi-model fusion
Shiyang Yan, Yu Zheng, Yixuan Chen, et al.
In this paper, a combinatorial algorithm-based visibility prediction method is proposed for improving the accuracy of visibility prediction. Firstly, four algorithms, namely support vector machine, kernel extreme learning machine, random forest and RBF neural network, are used as the basis functions for prediction, then the objective function of the combined prediction is constructed, the cuckoo search is used to optimise the calculation of the weighting coefficients of the combined prediction, and finally the combined prediction results are obtained. The experimental results show that the combined prediction algorithm proposed in this paper can effectively improve the accuracy of visibility prediction, and has certain application and research value.
Application of sigmoidal functions based on atomic functions in a machine learning problem of the analysis of biomedical and biometric data
Natalia Konnova, Mikhail Basarab
The paper proposes sigmoidal activation functions based on atomic functions. The properties of atomic functions are described, which allow them to satisfy the conditions for transfer functions in artificial neural networks. The applied problems of using the presented functions are considered on the example of problems of analysis of biophysical signals of cardiac activity. The results obtained using the constructed classifiers using various architectures of neural networks, including MLP, RNN, LSTM, GRU, CNN networks, are presented. The efficiency of using atomic functions in the constructed neural networks on the examples of the problem of automated diagnostics of pathologies based on seismocardiography data and biometric authentication by heart rate was determined using metrics of accuracy, recall, precision, sensitivity, specificity, and F-measure.
Formulation of precise short neural network code
Syeda Sadia Rubab, Ma Tian, Rehmat Bashir, et al.
In Deep Learning Artificial Neural Network is a comprehensive and mathematical way of explaining a simple 2-layered neural network by coding one from scratch in python and behind the scenes of such popular algorithms. First, we provided with a rationale view of the study behind the elaboration of these algorithms and mathematical intuition behind them. Then, we dived into the coding of neural network mixing python lines of code and mathematical equations. In this research, we created a two-layered neural network with a hidden layer to minimize time and effort resulting with more accurate and efficient output, but the idea remains the same for more than two-layered neural networks. Finally, we worked on how we could publicize our model and make it more adaptable for solving complex real-life issues and our experimental results showed that our 2-layered network model could achieve more accuracy and reliable measures as compared to other network methods for solving such complex problems.
The critical effect of diffusion coefficients and boundary data on the global exponential stabilization of delayed Markovian jump neural control systems
Puchen Liu, Yan Jin, Guoyan Cao
By employing Friedrichs’ inequality and constructing new Lyapunov-Krasovskii functional, we have investigated the stabilization of delayed reaction-diffusion control systems with Markovian jump parameters and Robin boundary conditions. Different from existing traditional results, our easy-to-test control design on global exponential stabilization in the mean square is based also on diffusion coefficients and boundary conditions, which discloses the crucial role of the diffusion effect and boundary information in the stabilization of neural control systems. The model here includes many models as special cases and the global exponential stabilization has a wider adaptive range. An example and its simulations are demonstrated to verify the effectiveness of theoretical results.
Synchronous analysis of a neural networks system
Lili Jia, Tao Jiang, Changyou Wang
In this article, we studies the synchronous problem of a fractional neural-networks. Based on the open-loop and adaptive feedback controls, we obtain the controller expression ensuring the projection synchronization for this models. Adopting V-function and applying comparison principle, a sufficient condition for the model to achieve projection synchronization is achieved.
Poetry can group: social relations among poets in Tang Dynasty based on Word2vec and social network
Yanbin Fang, Qinwei Li, Yuzhuo Liu, et al.
The study sought to dig deeply into the social relations of poets in Tang Dynasty through Tang poetry. Based on Word2vec, we trained the corpus of Tang poetry and generated the word vector, thus WMD was used to calculate the poetic similarity to reflect the poetic style; we use Gephi to construct the poets’ social relation network and use network analysis algorithms, including Community detection, Network shortcuts, etc to quantify the integral and partial attributes of the networks and to analyze the characteristics of "poetic social" and how the poet's own attributes affect the poet social on the macroscopic and microcosmic levels. Furthermore, poetic similarity were used to analyze the evolution characteristics of poetry styles under the influence of "poetic social". The study shows that there are some phenomena such as small groups and long tail effect in poets' social interaction; the social closeness of poets is related to the attributes of poets to varying extent; the social relations and poetic styles influence each other in development.
Residual neural network based joint transmit-receive antenna selection in MIMO system
Zhibin Guo, Jiaxin Cai
In large-scale MIMO system, as the number of antennas increases, the huge computational complexity makes traditional antenna selection algorithms impossible to effectively apply. This paper propose a joint transmitreceive antenna selection model based on ResNet. We utilize the optimal antenna selection algorithm to create labels for all channel matrices, which based on maximizing channel capacity criterion. Then using large-scale channel data to train a powerful residual neural network classifier. Consequently the trained model can classify the corresponding label for each channel matrix in the test set and select the optimal antenna subset. Experimental results show that the method can effectively decrease the number of antenna selection and its communication performance outperforms compared methods
A method for parameter optimization about the -SVR delay prediction
Liu Peiyan
The model of time delay prediction and the accuracy of parameters have great influence on power system network control,In this paper, a support vector machine (SVM) time delay modeling method based on bayesian evidence frame is proposed. This method treats the delay node as a "hidden layer", The time delay prediction model is established by using SVM regression theory. The gaussian kernel function is used in this model, and the parameters are optimized by using three criteria of bayesian evidence framework. This method can change the model structure, identify and optimize the parameters flexibly. In this paper, the model modeling and parameter optimization of delay prediction are completed.
Data Science and Artificial Intelligence Applications
icon_mobile_dropdown
Data-driven mining method for adjustable capacity of electric boiler with heat reservoir
Yong Sun, Yuhang Qiu, Baoju Li, et al.
In order to solve the problems of insufficient new energy consumption and serious abandonment of wind and light in the three north areas of China, the electric boiler is installed in the power grid area where the wind power consumption is insufficient, and the electric boiler is used to absorb wind power on the spot. In this paper, the heating demand and the maximum heat storage capacity of the electric boiler with heat reservoir are reversely deduced based on the user's electricity consumption data for the first time. According to the time of use electricity price in Jilin Province, the bubble sorting principle is used to tap the adjustable potential of the electric boiler with heat reservoir, and the peak power consumption is shifted to the low, so as to reduce the user's electricity charge and absorb the abandoned wind during the low period of the power grid, it plays the role of cutting peak and filling valley and stabilizing power grid. Compared with the brute force exhaustive method, this method greatly reduces the computational complexity. Compared with the method relying on the parameters of electric boiler and environmental parameters, this method greatly reduces the data which is needed to be collected, and reduces the evaluation errors caused by the performance degradation of electric boiler.
Sound field control in surround screen speaker array by COSH and SINC algorithms
Rui Huan, Xi Zhao
This paper proposes a method to solve the problem that the sound reproduction system cannot work when the movie screen is made by a sound-proof material such as LED. It is demonstrated in an array of 192 speakers to surround a screen for sound reproduction, called surround screen speaker array. The speaker array is built in an actual cinema. The sound field control algorithms are implemented by mixers. In order to improve the uneven sound field distribution and sound field aliasing caused by the speaker array, two algorithms COSH and SINC are used in this paper. A new control algorithm is proposed and demonstrated to improve the uniformity of the sound field distribution and reduce the sound field interference.
Research on innovative methods based on clustering and case-based reasoning
Innovation is one of the decisive factors for the prosperity of an enterprise. The Theory of Inventive Problem Solving (TRIZ) is an effective method to promote system innovative design. However, TRIZ solution is a highly abstract conceptual solution, and its transformation to a specific solution requires a large amount of rich knowledge from different engineering fields. Designers themselves cannot master all the supporting knowledge required for problem solving, which increases the difficulty of TRIZ in practical application. In order to overcome these difficulties, this paper presents a method synthesizing TRIZ, CBR and GHSOM, in which TRIZ can systematically generate innovative solutions, CBR makes designers take full advantage of past experience and knowledge to transform abstract TRIZ solutions into solutions to specific design problems, and GHSOM as a clustering tool expands the scope of knowledge utilization and promote the diversification of design solutions. Finally, an example is given to illustrate the method.
An overview of constructing geometric models of buildings using point clouds
Yuanzhi Cai, Lei Fan, Cheng Zhang
Building Information Modelling (BIM) and the associated numerical simulations have widely been recognised as useful components of building design support tools, and are readily applied in the design and the construction of new buildings. Meanwhile, their applications in existing buildings are often constrained by the lack of as-built geometric models, the generation of which often depends on two aspects: the spatial data (e.g., point cloud data) representing the building of interest and the means of constructing geometric models using the spatial data. The former can readily be obtained using laser scanning techniques. However, the generation of geometric models is mainly a manual process in practice, which is often time consuming. To improve the efficiency, extensive research has been carried out to investigate approaches for automating the geometric modelling process and improving the quality of the models obtained. This article provides a review of the approaches and the tools available in the literature for generating geometric models of buildings using point cloud data. The strengths and weaknesses of those approaches and the potential benefits of combining some of them are critically discussed.
Theoretical research and improvement of knowledge conversion (Withdrawal Notice)
Rongying Zhao, Zhuozhu Liu, Junling Wang, et al.
This paper reconsiders the SECI model proposed by Ikujiro Nonaka, and points out the inconformity between theory and practice in the original model when knowledge conversion is based on the theoretical basis and the actual situation, that is, socialization or combination can not be realized directly, but can only be realized indirectly through the combination of internalization and externalization. On this basis, the model is improved, and the spatial vector model of knowledge conversion is proposed. The SECI model is innovated, which provides theoretical support for the subsequent research on knowledge conversion. Through the establishment of knowledge conversion model in the spatial vector coordinate system, the specific transformation mode between explicit knowledge and tacit knowledge is embodied, and the inner conversion mode of combinatorial and socialized is refined. Through the study of knowledge conversion model, this paper holds that there are two direct ways of knowledge conversion, namely externalization and internalization.
Informetric analysis of researches on artificial intelligence in anthropology
Qiong Lan, Wenda Du, Fengjiao Xu
In the 21st century, one of the most important technologies must be artificial intelligence. It may bring us a whole new kind of people and technology, people and machines, or changes between people. So anthropological research is closely related to artificial intelligence research. This paper makes an infometric analysis to study 72 papers indexed in Thomson Reuters’s Web of Science by the perspectives of categories of researches, high frequency keywords, keywords co-occurrence and classifications of research topics. We uses VOSviewer to make a keywords co-occurrence map of the intersection of AI and anthropology which showed that artificial intelligence and interaction of anthropology focuses on the human spirit, body, technology improvement, the challenge of the moral and ethical aspects of the application, research is focused on the theory research and applied research. Researches of AI in anthropology mainly include two parts that the theoretical researches and the applied researches. The theoretical researches focuses on the relationship between human and AI, featuring the development of human beings in the future. The applied researches mainly includes the following three aspects: The application of AI in academic research, the application of AI in institutional service and the application of AI in people's daily life.
Research on application of artificial intelligence in movie industry
Wenda Du, Qing Han
The development of artificial intelligence has given birth to a series of innovations in the film industry. This article systematically summarizes these innovations into three parts; movie recommendation, movie distribution and audiovisual language creation. In the first part, the paper lists a series of movie recommendation systems based on AI technology. These systems can help users quickly find the movie resources they are interested in among the massive movie resources. In the second part, the article lists a series of cases of film production using AI, including analyzing film projects, estimating film box office, analyzing film reviews, and writing abstracts. In the third part, the article lists several cases about audio-visual language using AI systems,including classifying films and the aesthetic feature, analyzing image features through AI, changing the appearance and age of actors, and creating visual characters.
Research and application on virtual metering model for production of ESP well based on working performance curve
Mingyue Cui, Gang Feng, Junzheng Yang, et al.
Aiming at the problems of scattered oil wells and high production measurement cost of single well in Halfaya oilfield, in order to timely grasp the change trend of single well production and improve the intelligent management level of oil well , the virtual metering method of liquid production of Electric submersible pump (ESP) well in Halfaya oilfield is studied. Starting from the working performance curve of ESP well and considering the influence of gas, viscosity and water content in fluid on pump operation performance, this paper calculates and corrects the working performance curve of ESP well by using the least square curve and polynomial fitting method, and constructs the virtual metering mathematical model of ESP well production based on real-time monitoring data. The comparison and verification of field measurement shows that the virtual measurement model of ESP well production in this paper is simple and practical, and can realize the on-line measurement of ESP well production. The calculated results are in good agreement with the measured results, and the maximum relative error is less than 5%, which provides a basis for oil well dynamic analysis and working system optimization in Halfaya oilfield.
Preliminary exploration on digital transformation of Xinjiang oilfield
Hui Zhang, Chuanshuai Zuo, Zunchao Wei, et al.
In view of the fact that the informatization construction of Xinjiang Oilfield has entered the stage of intelligent construction, but there is still a certain gap in the digital transformation practice of advanced oil enterprises at home and abroad, it is proposed to build a digital technology covering oil and gas exploration, development and production, collaborative research The construction scheme of intelligent application ecology in business fields such as operation and management, so as to reduce the cost of oil and gas development and management, improve the efficiency of oil and gas development, and help the high-quality development of oil and gas companies. Taking the digital transformation of Xinjiang Oilfield as an example, this paper introduces the current situation of Internet of things construction in Xinjiang Oilfield. By combing the field production business process, combined with the development of big data and artificial intelligence in the field of oil and gas industry, this paper puts forward the scene construction of intelligent operation field, which provides guiding significance for the digital transformation of Xinjiang Oilfield.