A CNN-Based Single-Stage Occlusion Real-Time Target Detection Method

Aiming at the problem of low accuracy of traditional target detection methods for target detection in endoscopes in substation environments, a CNN-based real-time detection method for masked targets is proposed. The method adopts the overall design of backbone network, detection network and algorithmic parameter optimisation method, completes the model training on the self-constructed occlusion target dataset, and adopts the multi-scale perception method for target detection. The HNM algorithm is used to screen positive and negative samples during the training process, and the NMS algorithm is used to post-process the prediction results during the detection process to improve the detection efficiency. After experimental validation, the obtained model has the multi-class average predicted value (mAP) of the dataset. It has general advantages over traditional target detection methods. The detection time of a single target on FDDB dataset is 39 ms, which can meet the need of real-time target detection. In addition, the project team has successfully deployed the method into substations and put it into use in many places in Beijing, which is important for achieving the anomaly of occlusion target detection.

Application of Smith Predictor Based on Single Neural Network in Cold Rolling Shape Control

Flatness is one of the most important criterion factors to evaluate the quality of the steel strip. To improve the strip＇ s flatness quality, the most frequently used methodology is to employ the closed-loop automatic shape control system. However, in the shape control system, the shape-meter is always installed at the down way of the exit of the cold rolling mill and can not sense the changes of the strip flatness in the rolling gap directly. This kind of installation results in the delay of the feedback in the control system. Therefore, the stability and response performance of the system are strongly affected by the delay. At present, there is still no mature way to design controllers for systems with time delay. Although the conventional PID controller used in most practical applications has the capability to compensate the delay, the effect of the compensation is limited, especially for the systems with long time delay. Smith predictor, as a compensator for solving this problem, is now widely used in industry systems. However, the request of highly precise model of the system and the poor adaptive performance to the changes of related parameters limit the application of the Smith predictor in practice. In order to overcome the drawbacks of the Smith predictor, a new Smith predictor based on single neural network PID （SNN-PID） is proposed. Because the single neural network is employed into the Smith predictor to improve the controller＇s self-adaptability, the adaptive capability to the varying parameters of the system is improved. Meanwhile, for the purpose of solving the problems such as time-consuming and complicated calculation of the neural networks in real time, the learning coefficient of neural network is divided into several stages as usually done in expert control system. Therefore, the control system can obtain fast response due to the improved calculation speed of the neural networks. In order to validate the performance of the proposed controller, the experiment is conducted on the shape control system in a 300 mm four-high reversing cold rolling mill. The experimental results show that the SNN-PID with Smith predictor controller can effectively compensate the delay effects and achieve better control performance than the conventional PID controller.

Single Phase Induction Motor Drive with Restrained Speed and Torque Ripples Using Neural Network Predictive Controller

In industrial drives, electric motors are extensively utilized to impart motion control and induction motors are the most familiar drive at present due to its extensive performance characteristic similar with that of DC drives. Precise control of drives is the main attribute in industries to optimize the performance and to increase its production rate. In motion control, the major considerations are the torque and speed ripples. Design of controllers has become increasingly complex to such systems for better management of energy and raw materials to attain optimal performance. Meager parameter appraisal results are unsuitable, leading to unstable operation. The rapid intensification of digital computer revolutionizes to practice precise control and allows implementation of advanced control strategy to extremely multifaceted systems. To solve complex control problems, model predictive control is an authoritative scheme, which exploits an explicit model of the process to be controlled. This paper presents a predictive control strategy by a neural network predictive controller based single phase induction motor drive to minimize the speed and torque ripples. The proposed method exhibits better performance than the conventional controller and validity of the proposed method is verified by the simulation results using MATLAB software.

R2N: A Novel Deep Learning Architecture for Rain Removal from Single Image

Visual degradation of captured images caused by rainy streaks under rainy weather can adversely affect the performance of many open-air vision systems.Hence,it is necessary to address the problem of eliminating rain streaks from the individual rainy image.In this work,a deep convolution neural network(CNN)based method is introduced,called Rain-Removal Net(R2N),to solve the single image de-raining issue.Firstly,we decomposed the rainy image into its high-frequency detail layer and lowfrequency base layer.Then,we used the high-frequency detail layer to input the carefully designed CNN architecture to learn the mapping between it and its corresponding derained high-frequency detail layer.The CNN architecture consists of four convolution layers and four deconvolution layers,as well as three skip connections.The experiments on synthetic and real-world rainy images show that the performance of our architecture outperforms the compared state-of-the-art de-raining models with respects to the quality of de-rained images and computing efficiency.

Improved artificial neural network method for predicting photovoltaic output performance

To ensure the safety and stability of power grids with photovoltaic(PV)gen eration integrati on,it is necessary to predict the output perform a nee of PV modules un der varyi ng operating con ditions.In this paper,an improved artificial neural network(ANN)method is proposed to predict the electrical characteristics of a PV module by combining several neural networks under different environmental conditions.To study the dependenee of the output performance on the solar irradianee and temperature,the proposed neural network model is composed of four neural networks,it called multineural network(MANN).Each neural network consists of three layers,in which the input is solar radiation,and the module temperature and output are five physical parameters of the single diode model.The experimental data were divided into four groups and used for training the neural networks.The electrical properties of PV modules,including l-V curves,PV curves,and normalized root mean square error,were obtained and discussed.The effectiveness and accuracy of this method is verified by the experimental data for d iff ere nt types of PV modules.Compared with the traditional single-ANN(SANN)method,the proposed method shows be社er accuracy under different operating conditions.

Expert control strategy using neural networks for electrolytic zinc process

The most important parameters which control the electrolytic process are the concentrations of zinc and sulfuric acid in the electrolyte. An expert control strategy for determining and tracking the optimal concentrations was proposed, which uses neural networks, rule models and a single loop control scheme. First, the process was described and the strategy that features an expert controller and three single loop controllers was explained. Next, neural networks and rule models were constructed based on statistical data and empirical knowledge on the process. Then, the expert controller for determining the optimal concentrations was designed through a combination of the neural networks and rule models. The three single loop controllers used the PI algorithm to track the optimal concentrations. Finally, the implementation of the proposed strategy were presented. The run results show that the strategy provides not only high purity metallic zinc, but also significant economic benefits.

Neural network modeling and intelligent control of FCAW penetration

The neural network modeling of FCAW penetration is researched in this paper, molten pool image is acquired by CCD, and preweld gap is gotten from laser vision system, the weld penetration is estimated according to the information include welding current, welding voltage, weld width, molten pool half length and gap width. The training samples of network can be partially gotten by numerical simulation. Single neuron self-tuning PID weld penetration controller is designed, and improved Hebb learning algorithm is applied for weights adjusting. Welding current is adjusted to make the weld penetration stable. The results of experiment with various cross-section and preweld gap workpiece show that this system is suitable to molten pool control.

CT reconstruction from a single X-ray image for a particular patient via progressive learning

Computed tomography(CT)has enjoyed widespread applications,especially in the assistance of clinical diagnosis and treatment.However,fast CT imaging is not available for guiding adaptive precise radiotherapy in the current radiation treatment process because the conventional CT reconstruction requires numerous projections and rich computing resources.This paper mainly studies the challenging task of 3 D CT reconstruction from a single 2 D X-ray image of a particular patient,which enables fast CT imaging during radiotherapy.It is widely known that the transformation from a 2 D projection to a 3 D volumetric CT image is a highly nonlinear mapping problem.In this paper,we propose a progressive learning framework to facilitate 2 D-to-3 D mapping.The proposed network starts training from low resolution and then adds new layers to learn increasing high-resolution details as the training progresses.In addition,by bridging the distribution gap between an X-ray image and a CT image with a novel attention-based 2 D-to-3 D feature transform module and an adaptive instance normalization layer,our network obtains enhanced performance in recovering a 3 D CT volume from a single X-ray image.We demonstrate the effectiveness of our approach on a ten-phase 4 D CT dataset including 20 different patients created from a public medical database and show its outperformance over some baseline methods in image quality and structure preservation,achieving a PSNR value of 22.76±0.708 dB and FSIM value of 0.871±0.012 with the ground truth as a reference.This method may promote the application of CT imaging in adaptive radiotherapy and provide image guidance for interventional surgery.

THE CONTROL, DRIVE AND DISPLAY CIRCUITS ANDDEVICES IN OPTO-ELECTRONIC 2-D PROGRAMMABLE NEURAL NETWORK

In an optoelectronic 2-D programmable neural network system, optical data need to be transferred and feedback with high speed. This paper presents the design and implementation of the interface circuit and its software.

Effective prediction of DEA model by neural network

In this paper,a fast neural network model for the forecasting of effective points by DEA model is proposed,which is based on the SPDS training algorithm.The SPDS training algorithm overcomes the drawbacks of slow convergent speed and partially minimum result for BP algorithm.Its training speed is much faster and its forecasting precision is much better than those of BP algorithm.By numeric examples,it is showed that adopting the neural network model in the forecasting of effective points by DEA model is valid.

Design of an Automated Sorting System for Apples Based on Single Chip Microcomputer

The grading judgment for apples is related to a variety of factors including,size,shape,color,texture,and scars.Traditional manual sorting methods are time consuming and labor intensive.In addition,the accuracy of the method is easily subjective,not repeatable,error-prone,and affected by the sorting environment.This paper presents a complete and automated grading system for apples.The system uses a single-chip microcomputer as the controller of the system,and a PC as the graphics processing unit.It also includes a conveyor,drive motor,frequency converter for motor control,photoelectric sensors,air compressor,and air jets for ejecting the graded apples.The classification algorithm is implemented by using a convolutional neural network(CNN).In order to eliminate contact damage of apples,the system specifically uses air jets as actuators to eject the graded apples into the corresponding bins.At the same time,in order to ensure that an apple triggers the correct ejecting actuator,this paper designs a jet controller with proper logic.

Neural Network Based on SET Inverter Structures: Neuro-Inspired Memory

This paper presents a basic block for building large-scale single-electron neural networks. This macro block is completely composed of SET inverter circuits. We present and discuss the basic parts of this device. The full design and simulation results were done using MATLAB and SIMON, which are a single-electron tunnel device and circuit simulator based on a Monte Carlo method. Special measures had to be taken in order to simulate this circuit correctly in SIMON and compare results with those of SPICE simulation done before. Moreover, we study part of the network as a memory cell with the idea of combining the extremely low-power properties of the SET and the compact design.

图计算体系结构和系统软件关键技术综述

图计算作为分析事物之间关联关系的重要工具,近年来已成为各国政府及公司争夺的关键技术.学术界和工业界在图计算体系结构和系统软件关键技术方面取得了一定进展.然而,现实场景图计算大多具有动态变化、应用需求复杂多样等特征.这给图计算在基础理论、体系架构和系统软件关键技术方面提出了新的需求,同时也带来了新的挑战.为应对这些挑战,科研人员提出了一系列图计算系统或图计算加速器,通过高性能计算、并行计算等技术来优化图计算过程.综述国内外图计算体系结构和系统软件关键技术的研究发展现状,对国内外研究的最新进展进行归纳、比较和分析,并结合国家发展战略和重大应用需求,选取与我国国计民生密切相关的领域,从典型应用分析总结图计算相关技术的行业进展.最后,就未来的技术挑战和研究方向进行展望.

基于曲线拟合和神经网络的独头巷道CO浓度预测研究

为了准确预测矿山独头巷道CO浓度,基于云南老厂锡矿1800运输巷甩车场独头巷道掘进工作面CO浓度监测数据,运用MATLAB曲线拟合工具箱对该独头巷道中CO浓度随时间的变化情况进行曲线拟合,建立了该矿山独头巷道中CO浓度随时间变化的数学模型。通过该模型得到该独头巷道中CO浓度值达到安全规程要求所需的时间。然后,运用卷积神经网络时间序列预测模型(CNN模型)和BP神经网络时间序列预测模型(BP模型)对独头巷道CO浓度进行预测,并比较评价指标R2和RMSE。结果表明:BP神经网络时间序列预测模型对该独头巷道CO浓度的预测效果更好,为该矿山独头巷道CO浓度值的监测和控制提供了准确可靠的理论依据。

适用于图像超分辨率的多路径融合增强网络

卷积神经网络(Convolutional Neural Network,CNN)在单幅图像的超分辨率重建方面表现出了非常强大的能力,相比传统方法有着明显的改进。然而,尽管这些方法非常成功,但是由于需要大量的计算资源,直接应用于一些边缘设备并不现实。为了解决该问题,设计了一种轻量级的图像超分辨率重建网络——多路径融合增强网络(Multi-path Fusion Enhancement Network,MFEN)。具体来说,提出了一个新颖的融合注意力增强模块(Fusion Attention Enhancement Block,FAEB)作为多路径融合增强网络的主要构建模块。融合注意力增强模块由一条主干分支和两条层级分支构成:主干分支由堆叠的增强像素注意力模块组成,负责对特征图实现深度特征学习;层级分支则负责提取并融合不同大小感受野的特征图,从而实现多尺度特征学习。层级分支的融合方式则是以相邻的增强像素注意力模块输出为分支输入,通过自适应注意力模块(Self-Adaptive Attention Module,SAAM)来动态地增强不同大小感受野特征的融合程度,进一步补全特征信息,从而实现更全面、更精准的特征学习。大量实验表明,该多路径融合增强网络在基准测试集上具有更高的准确性。

基于变分模态分解与空洞卷积神经网络的配电网故障选线方法

小电流接地系统发生单相接地故障时,零序电流故障特征微弱且繁杂多变,传统选线方法可靠性有待提高。提出了一种基于变分模态分解(variational mode decomposition,VMD)与空洞卷积神经网络的配电网故障选线方法。首先,分析配电网健全线路和故障线路的电气特征,采用零序电流作为故障特征信号,为选线模型的输入量提供理论依据;其次,通过变分模态分解把零序电流序列分成不同频率的固有模态函数,提高故障信号特征的平稳性和差异性;然后,采用空洞卷积神经网络作为选线网络,以增大卷积操作感受野的方式增强模型的自适应分类能力;最后,在MATLAB/Simulink中构建10kV配电网进行算例分析,结果表明,该方法在不同故障场景条件下均有较高的选线效果,验证了所提方法的鲁棒性与准确性。

基于深度学习的木材缺陷智能检测的研究进展与展望

木材作为天然生物材料很容易受到内外界影响从而产生不符合人们生产需求的缺陷,人们为了准确高效的识别木材缺陷进行了大量的研究。本文对近年来基于深度学习的木材缺陷检测技术进行梳理,根据使用方法的侧重点不同将其分类,并针对典型方法加以细分归类和对比分析,总结了每种方法的优缺点及其应用面。此外,提出了基于深度学习的木材缺陷检测技术目前所存在的难点与所陷困境。

Realize ultralow-energy-consumption photo-synaptic device based on a single(Al,Ga)N nanowire for neuromorphic computing

The rapid development of artificial intelligence poses an urgent need for low-energy-consumption and small-sized artificial photonic synapses.Here,it is pretty novel to demonstrate a light-stimulated synaptic device based on a single(Al,Ga)N nanowire successfully.Thanks to the presence of vacancy defects in the single nanowire,the artificial synaptic device can simulate multiple functions of biological synapses under stimulation of both 310 and 365 nm light photons,including paired-pulse facilitation,spike timing dependent plasticity,and memory learning capabilities.The energy consumption of artificial synaptic device can be reduced as little as 5.58×10^(-13) J,which is close to that of the biological synapse in human brain.Furthermore,the synaptic device is demonstrated to have the high stability for both long-time stimulation and long-time storage.Based on the experimental conductance of long-term potentiation and long-term depression,the simulated three-layer neural network can achieve a high recognition rate of 92%after only 10 training epochs.With a brain-like behavior,the single-nanowire-based synaptic devices can promote the development of visual neuromorphic computing technology and artificial intelligence systems requiring ultralow energy consumption.

足式机器人腿部关节改进单神经网络PID控制算法研究

为了满足液压足式机器人在复杂环境中实现精确、快速的腿部关节控制需求,把单神经网络PID能够实时调节参数的优点运用到足式机器人液压机械腿关节的控制中,在单神经网络PID的基础上增加机械腿关节的位置和速度控制算法,形成改进单神经网络PID,实现了对神经元比例参数自调整、PID参数的自整定,能够较好地适应内、外参数的变化,增强了腿部关节的快速性、精确性。在Simulink中进行建模仿真以及在设计的以STM32为中央处理芯片的控制平台上进行实验测试,结果表明:改进单神经网络PID在足式液压机器人的腿部关节控制中具有响应速度快、超调量小、控制精度高、鲁棒性强等优点。