With the development of Internet of Things (IoTs), the number of sensor nodes is growing rapidly. These sensors are usually passive or supplied by batteries, and are usually a mixed-signal circuit. Analog to digital converter (ADC) is a core element in the sensor, and the power consumption of occu-pies a considerable part of the whole sensor. SAR ADC is a good candidate for the sensor due to its good energy-efficiency, medium resolution and speed. As the key part of SAR ADC, digital-to-analog converter (DAC) dom-inates the power consumption of the SAR ADC when dynamic comparator is employed. In order to improve the energy efficiency of the DAC, this paper proposes energy-efficient DAC scheme with based on unit capacitor switch-ing. By employing a capacitor-splitting structure and introducing a third volt-age reference Vq equal to a quarter of the voltage reference Vref, the unit ca-pacitor can be employed to generate the last bit, which in turn reduces the DAC area. Simulation results show that the proposed scheme reduces the switching energy by 99.03% and the DAC area by 87.5% compared to the conventional SAR ADC structure, which achieves good energy-efficiency and area-efficiency.
Authors: Liangbo Xie (School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China), Yan Ren (School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, , Chongqing, China),
Hide Authors & Abstract

Accurate meteorological observation relies heavily on the proper and precise working of meteorological facilities. Nevertheless, a big portion of meteorological facilities are deployed in outdoor environments hardly within the reach of convenient monitoring and reliable networking infrastructure, hence the surveillance challenge. Given such an infrastructure-less/-poor environment (deserts, oceans, etc.) for meteorological facilities, Ad Hoc nodes are possible candidates for mobile surveillance. However, pure Ad Hoc networking without a logical centric node can barely provide consistent collaboration between mobile nodes during monitoring. This paper proposes a tunnelled overlay structure that bridges the Ad Hoc protocol stack and the SDN (Software-Defined Networking) protocol stack based on network virtualization techniques, so that robust distributed Ad Hoc mobile nodes are grouped in the form of an SDN-governed swarm to conduct the mobile surveillance task with joint efforts under the consistent control of the centric SDN controller. In addition, mobile nodes are equipped with TensorFlow-based image recognition feature, capable of transmitting the recognized results of harmful creatures that might cause facility damages, to suggest proper protection measures, control/avoid facility loss, etc. Experiments on the prototype SDN-governed Ad Hoc swarm are carried out in a real-world university campus meteorological station to demonstrate its feasibility and functionalities.
Authors: Xi Chen (Southwest Minzu University), Tao Wu (Chengdu University of Information Technology), Ying Tan (Southwest Minzu University),
Hide Authors & Abstract

In the hardware Trojan detection method, destructive reverse engineering can most precisely restore the original circuit of the chip to be detected, but this method is a huge amount of work, high cost, long life cycle. In this paper, we proposed a multi-featured GEP technology, non-destructive reverse engineering of the chip using various data obtained from bypass detection, in order to restore the actual circuit of the hardware, or at least find out the unknown circuit design.
Authors: Huan Zhang (Sichuan University), Jiliu Zhou (Sichuan University), Xi Wu (Chengdu University of Information Technology), Yi Zhang (Sichuan University),
Hide Authors & Abstract

In automatic navigation robots, robotic autonomous positioning is one of the most difficult challenges. Simultaneous Localization and Mapping (SLAM) tech-nology can incrementally construct a map of the robot's moving path in an un-known environment while estimating the position of the robot in the map, provid-ing an effective solution for robots to fully navigate autonomously. The camera can obtain corresponding two-dimensional digital images from the real three-dimensional world. These images contain very rich color, texture information and highly recognizable features, which provide indispensable information for robots to understand and recognize the environment based on the ability to autonomous-ly explore the unknown environment. Therefore, more and more researchers use cameras to solve SLAM problems, also known as visual SLAM. Visual SLAM needs to process a large number of image data collected by the camera, which has high performance requirements for computing hardware, and thus its application on embedded mobile platforms is greatly limited. In this re-gard, this paper uses embedded hardware equipped with embedded GPU, com-bines CUDA-based GPU parallel computing and visual SLAM algorithm, final-ly, designs a parallelization scheme based on embedded GPU.
Authors: Tianji Ma (Chengdu University of Information Technology), Nanyang Bai (Chengdu University of Information Technology), Wentao Shi (Chengdu University of Information Technology), Lutao Wang (Chengdu University of Information Technology), Tao Wu (Chengdu University of Information Technology),
Hide Authors & Abstract

In recent years, deep learning-based image extrapolation has achieved remarkable improvements. Image extrapolation utilizes the structural and semantic information from the known area of an image to extrapolate the unknown area. In addition, these extrapolative parts not only maintain the consistency of spatial information and structural information with the known area, but also achieve a clear, beautiful, natural and harmonious visual effect. In view of the shortcomings of traditional image extrapolation methods, this paper proposes an image extrapolation method which is based on perceptual loss and style loss. In the paper, we use the perceptual loss and style loss to restrain the generation of the texture and style of images, which improves the distorted and fuzzy structure generated by traditional methods. The perceptual loss and style loss capture the semantic information and the overall style of the known area respectively, which is helpful for the network to grasp the texture and style of images. The experiments on the Places2 and Paris StreetView dataset show that our approach could produce better results.
Authors: Yongpeng Ren (Chengdu University of Information Technology), Xian Zhang (Chengdu University of Information Technology), Hongping Ren (Chengdu University of Information Technology), Lutao Wang (Chengdu University of Information Technology), Guanyao Huang (Chengdu Shengdaren Technology Co. Ltd), Xiaojie Li (Chengdu University of Information Technology), Taisong Xiong (Chengdu University of Information Technology),
Hide Authors & Abstract

In this paper, we consider an impulsive mixture noise process, which is commonly encountered in applications such as multiuser radar communications and astrophysical imaging. The mixture process is in the time domain, the probability density function (PDF) of which corresponds to the convolution of the components' PDFs. In this work, we concentrate on the additive mixture of Gaussian and Cauchy variables, whose PDF is their convolution, leading to a Voigt profile. Due to the complicated nature of the PDF, classical methods such as maximum likelihood estimator may be analytically not tractable. Therefore, to estimate parameters under such noise, we propose using a Markov chain Monte Carlo method, namely, the Metropolis-Hastings algorithm. For illustration, we address the frequency estimation of a single sinusoid embedded in the Cauchy-Gaussian mixture noise. Simulation results demonstrate that the mean square error performance of the proposed algorithm can attain the Cram\'{e}r-Rao lower bound.
Authors: Yuan Chen (University of Science and Technology Beijing), Dingfan Zhang (Beihang University), Longting Huang (Wuhan University of Technology),
Hide Authors & Abstract

The modulation measurement profilometry encoding the spatial distribution information of specimen surface into the fringe defocus can realize the reconstruction of the specimen with complex surface shape. For this technique, the imaging axis of the CCD camera is coaxial with the projecting direction thanks to the application of a beam splitter mounting in the projection optical path. Without doing the phase unwrapping operation, it can accomplish shadow-free measurement for the specimen by extracting modulation values of the fringe pattern. The paper makes a comparison of the modulation retrieval in the conditions of fringe patterns with different surface profiles. Two Fourier transform methods are implemented in our computer simulation and practical experiment to show their performance in demodulating the modulation information from fringe patterns in optical 3D shape measurement.
Authors: Min Zhong (Chengdu University of Information Technology), Feng Chen (Chengdu University of Information Technology), Chao Xiao (Chengdu University of Information Technology), Peng Duan (Chengdu University of Information Technology), Min Li (Chengdu University of Information Technology), Wenzao Li (Chengdu University of Information Technology),
Hide Authors & Abstract

The main goal of this paper is to investigate the problems of the uniqueness of equilibrium and the global µ-stability for the QVNN (quaternion-valued neural network) with leaky constant delay, non-differentiable discrete time-varying delay, distributed constant delay, which is closer to practical application than the QVNN with differentiable time-varying delay. Firstly, we discuss the QVNN as entirety, and prove the equilibrium of the QVNN is unique by using Homeomorphism mapping theorem and quaternion-valued linear matrix inequality. Then a new Lyapunov-Krasovskii functional is derived from the delayed state. The sufficient condition of the global µ-stability is given, while appraising the derivative of the Lyapunov-Krasovskii functional and quaternion-valued linear matrix inequality, this result is new and different from the approaches in available literatures. A quaternion-valued numerical example is presented to illustrate these results.
Authors: Hongying Qin (Leshan Normal University), Zhenhao Chen (University of Electronic Science and Technology of China), Xiaomei Wang (University of Electronic Science and Technology of China), Guo Huang (Leshan Normal University),
Hide Authors & Abstract

IAA(Iterative Adaptive Approach) is a nonparametric algorithm which has high resolution in spectrum analysis. The basic idea of IAA is grid searching in $[0,2\pi)$ so the number of grid points determines the accuracy and complexity. In order to apply IAA in large data size, we consider shorten frequency range to decrease complexity accordingly. Being Limited by definition of weighting matrix, IAA cannot cut down range directly so we propose a new definition of signal and noise in order to estimate the spectrum of selective range.
Authors: Yuan Chen (University of Science and Technology Beijing), Longting Huang (Wuhan University of Technology),
Hide Authors & Abstract

In this paper, a mixture process is proposed for modelling impulsive noise, which is a sum of Cauchy and Gaussian random variables. The probability density function (PDF) of the mixture is referred to as the Voigt function derived from the convolution of the Cauchy and Gaussian PDFs. To determine the parameter of the constant model in the environment of the additive mixture noise, the maximum likelihood estimator is first developed. Since this method suffers from a complicated analytical form, an $M$-estimator with pseudo-Voigt function is also devised. In our study, we consider both scenarios of known and unknown density parameters. Simulation results show that the mean square error performance of both proposals can attain the Cram\'{e}r-Rao lower bound.
Authors: Yuan Chen (University of Science and Technology Beijing), Dingfan Zhang (Beihang University,Beijing,China), Longting Huang (Wuhan University of Technology),
Hide Authors & Abstract

In this work, the performance of four Machine Learning Algorithms (MLAs) applied to Cognitive Mobile Radio Networks (CMRNs) are analyzed. These algorithms are Coalition Game Theory (CGT), Naive Bayesian Classier (NBC), Support Vector Machine (SVM), and Decision Trees (DT). The numerical results of the performance analysis of these algorithms are presented based on two metrics. These metrics are commonly used in CMRNs which are Probability of Detection (Pd) and Probability of False Alarm (Pfa) against Signal-to-Noise Ratio (SNR). Furthermore, outcomes regarding the Classication Quality (CQ) and the simulation time are exposed. Theoretical and numerical results show that the SVM outperforms the rest of the algorithms in each of the metrics. The reasons behind this come from the SVM features, namely high precision, fast learning, and simplicity in the realization stage.
Authors: Pablo Palacios (University of Chile), Cesar Azurdia-Meza (University of Chile), Ivàn Sànchez (Universidad de las Amèricas), David Zabala-Blanco (Universidad Católica del Maule), Milton Roman (Universidad de Màlaga),
Hide Authors & Abstract

Recent studies attempt to construct complicated and redun-dant Convolutional Neural Networks (CNNs) to improve image classifi-cation performance. In this paper, instead of painstakingly designing aCNN’s architecture, we consider promoting classification performance byrevising CNN’s classification results. We therefore propose a novel im-age classification approach that Learns to Rectify Label (LRL) throughKernel Extreme Learning Machine (KELM). It includes two phases: (1)Pre classification, we put images into a trained CNN to generate corre-sponding incomplete labels. (2) Label Rectification, the incomplete labelsare rectified by the KELM’s high-dimensional mapping, so final classifi-cation results are acquired. Extensive experiments conducted on publicdatasets demonstrate the effectiveness of our method. At the meantime,our method has well generalizability that can be integrated with manypopular networks.
Authors: Qiang Cai (Beijing Technology and Business University), Fenghai Li (Beijing Technology and Business University), Haisheng Li (Beijing Technology and Business University), Jian Cao (Beijing Technology and Business University), Shanshan Li (Beijing Technology and Business University),
Hide Authors & Abstract

To enhance the edge and texture information of the image, enhance the contrast of the image effectively, and then improve the visual effect of the image, an image enhancement model based on contrast limited adaptive histogram equalization incorporating a fractional differential operator is proposed. The image enhancement model incorporates a fractional differential operator into the adaptive limited contrast image enhancement model, which can enhance the image contrast and effectively enhance the edge and texture details of the image simultaneously. Experimental results show that the proposed variable-order fractional differential contrast-limited adaptive histogram equalization image enhancement model can significantly improve the contrast of the image compared with the traditional fractional differential image enhancement model; additionally, it can effectively enhance the edge and texture details of the image compared with the traditional image enhancement model, which is based on statistical methods.
Authors: Guo Huang (Sichuan Province University Key Laboratory of Internet Natural Language Intelligent Pro-cessing, Le shan Normal University), Li Xu (Leshan Normal University), Li Chen (Sichuan Province University Key Laboratory of Internet Natural Language Intelligent Pro-cessing), Tao Men (Sichuan Province University Key Laboratory of Internet Natural Language Intelligent Pro-cessing), Ying Qing (Sichuan Province University Key Laboratory of Internet Natural Language Intelligent Pro-cessing), Qiong Zhang (Sichuan Province University Key Laboratory of Internet Natural Language Intelligent Pro-cessing),
Hide Authors & Abstract

Using one or several 2D images to get 3D face structure has always been one of the hot topics in computer vision, and it is a big challenge for all researchers. There are several problems, such as lack of training data and af-fected by the large pose. Traditional 3D face reconstruction，like 3DMM，over-rely on the accuracy of landmarks and the accuracy of face detector. Recent years, neural network application in face reconstruction, ac-cording to PRNet, we build a 2D representation called UV space, and using a simple network to regress it. In the loss function, we divided the face into 4 areas by facial landmarks, and give each area different weight. In this way, our method performs good results in both AFLW2000-3D and Florence da-taset. It can reconstructs 3D facial structure directly from a single 2D facial image.This is an end-to-end method, and our network is light-weighted.
Authors: Wentao Shi (Chengdu University of Information and Technology), Lutao Wang (Chengdu University of Information and Technology), Tianji Ma (Chengdu University of Information and Technology), Nanyang Bai (Chengdu University of Information and Technology),
Hide Authors & Abstract

Simultaneous localization and mapping (SLAM) is considered to be the basic ability of intelligent mobile robots. In the past few decades, thanks to community’s continuous and in-depth research on SLAM algorithms, the current SLAM algorithms have achieved good performance. But there are still some problems. For example, most SLAM algorithms have the assumption of a static environment, but in real life, most of the environment contains moving objects, so how to deal with the moving objects in the environment requires careful consideration. What's more, traditional geometric maps cannot specific environmental semantic information for mobile robots, so how to make robots truly understand the surrounding environment to complete some advanced tasks is also a difficult problem. In this paper, we design a scheme to improve the accuracy and robustness of SLAM in a dynamic environment. And we realize the perception of semantic information of objects in the environment through the object detection algorithm of deep learning neural network.
Authors: Nanyang Bai (Chengdu University of Information Technology), Tianji Ma (Chengdu University of Information Technology), Wentao Shi ( Chengdu University of Information Technology), Lutao Wang ( Chengdu University of Information Technology),
Hide Authors & Abstract

Sleep apnea is an important factor that affects human health. Tradi- tional approaches based on wearable devices or pressure sensor devices are too expensive to be suitable for daily use, which also don’t consider the impact on the breathing frequency when the human body turns over or gets up. In this paper, we propose a system based on WiFi to monitor sleep apnea state. Firstly, we use linear fitting to eliminate the phase errors of the receiving antennas, and wavelet transform to remove the noise of signal amplitude. Secondly, we combine the short-time Fourier transform and sliding window method to segment the signal. Finally, the features such as the variance of the phase difference between anten- nas are extracted, and the neural network model is built to identify apnea state to eliminate interference caused by changes in sleep postures. Experiment results show that the detection accuracy rate for apnea is over 95.6%. Our system can be a daily apnea monitoring approach and provide health reference for users.
Authors: Xiaolong Yang (Chongqing University of Posts and Telecommunications), Xin Yu (Chongqing University of Posts and Telecommunications), Liangbo Xie (School of Communication and Information Engineering, Chongqing University of Posts and Telecommunications, Chongqing 400065, China), Mu Zhou (CQUPT), Qing Jiang (Chongqing University of Posts and Telecommunications),
Hide Authors & Abstract

Generative adversarial networks (GAN) could synthesize semantically mean-ing data from standard signal distribution, which make it have considerable potential to alleviate data scarcity. In this paper, based on Evolutionary GAN, cardiac magnetic resonance images enhancement method is proposed to solve over-fitting problem caused by training convolution network with small dataset. The most optimal generator which consider the quality and di-versity of generated images simultaneously from many generator mutations is chosen. Meanwhile, to expand the whole training set distribution, we com-bine the linear interpolation of eigenvectors to synthesize new training sam-ples and synthesize related linear interpolation labels, which can make the discrete sample space become continuous to improve the smoothness be-tween domains. In this paper, the effectiveness of this method is verified by classification experiments, and the influence of the proportion of synthe-sized samples on the classification results of cardiac magnetic resonance im-ages is explored.
Authors: Ying Fu (Chengdu University of Information Technology), XueMin Gong (Chengdu University of Information Technology), Guang Yang (Chengdu University of Information Technology), JiLiu Zhou (Chengdu University of Information Technology),
Hide Authors & Abstract

Charging stations deployment is an important problem in Electric Vehicle (EV) networks. The distribution of EV is complicated in urban environments. Therefore, reasonable location deployment will avail to reduce construction costs and improve user experience. Aim to this, this paper comprehensively considers the cost of charging stations and the charging costs of EVs.Studied the charging station location, charging station capacity and charging station location optimization algorithm, proposed a method for estimating the optimal location and optimal capacity allocation of EV charging stations. Firstly, this paper uses the Voronoi diagram to divide the service range of the charging stations, then uses the differential evolution algorithm combined with the particle swarm optimization algorithm (DEIPSO) to solve the charging station location model, and finally consider the residence time of EV in the charging station, use queuing theory to solve the charging station capacity allocation model.The experimental results shows that DEIPSO can better jump out of the local optimum and achieve the global optimum; the proposed model can plan the charging station on the basis of fully considering the total charging costs of charging stations and EVs.
Authors: Lingling Yang (College of Communication Engineering, Chengdu University of Information Technology), Jiali Chen (College of Communication Engineering, Chengdu University of Information Technology), Wenzao Li (College of Communication Engineering, Chengdu University of Information Technology; School of Computing Science, Simon Fraser University.), Zhan Wen (College of Communication Engineering, Chengdu University of Information Technology),
Hide Authors & Abstract

Heterogeneous sensor networks, including water distribution systems and traffic monitoring systems, produce abundant time series data with an arbitrarily-high multivariate order for monitoring network dynamics and detecting events of interest. Nevertheless, errors and failures in the calibration, data storage or acquisition can occur on some of the sensors installed in those systems, producing missing and/or anomalous values. This work proposes a computational system, referred as AutoMTS, for the fully autonomous cleaning of multivariate time series data using strict quality criteria assessed against ground truth extracted from the targeted series data. The proposed methodology is parameter-free as it relies on robust principles for the assessment, hyperparameterization and selection of methods. AutoMTS coherently supports an extensive set state-of-the-art methods for (multivariate) time series imputation and outlier detection-and-treatment, considering both point and segment/serial occurrences. A comprehensive evaluation of AutoMTS is accomplished using heterogeneous sensors from two water distribution systems with varying sampling rates, water consumption patterns, and error profiles. Results confirm the relevance of the proposed AutoMTS system. AutoMTS is provided as an open-source tool available at https://github.com/RicardoFLNSousa/AutoMTS/tree/master .
Authors: Ricardo Sousa (CEMAT, INESC-ID and Instituto Superior Técnico, Universidade de Lisboa), Conceição Amado (CEMAT and Instituto Superior Técnico, Universidade de Lisboa), Rui Henriques (INESC-ID and Instituto Superior Técnico, Universidade de Lisboa),
Hide Authors & Abstract

Mobile edge computing provides a completely new technology, computational of-floading， that minimizes wireless communication latency by offloading compu-ting tasks on the mobile device to edge servers, but one problem is how to select and evaluate the offloading ratio of computations. In the paper, we propose a new computational offloading evaluation method, Computation Offloading Evaluation in Rayleigh (COER）, which is a method for estimating the task offloading ratio based on SNR. By studying the SNR spectrum of both Rayleigh Fading Channel (RFC) and Gaussian channel in 5G network, using the Monte Carlo method to statistically analyze the Signal-to-Noise Ratio (SNR) in the simulated RFC, and obtained the upper and lower limits of the SNR in the multipath fading channel. And based on the upper and lower limits of SNR, we get different transmission rates, which combined with the delay of the computationally intensive task during offloading, a more reasonable computation offloading ratio is obtained. After comparing the computation offloading ratio in the Gaussian channel, the experi-mental results show that the new calculation offloading ratio is more suitable for the real communication environment. This paper provides a new and more accu-rate computation offloading evaluation method for the delay minimization of mo-bile device in 5G networks.
Authors: Xudong Tang (Chengdu University of Information Technology), Chen Xue (Chengdu University of Information Technology), Xi Chen (Southwest Minzu University), Xi Wu (Chengdu University of Information Technology), Tao Wu (Chengdu University of Information Technology), Changming Zhao (Chengdu University of Information Technology),
Hide Authors & Abstract