基于高效Transformer的图像去雾研究

随着Transformer技术发展使其在计算机视觉(CV)领域中大放异彩,其构建全局依赖关系能力是卷积神经网络(CNN)所不具备的,而Transformer对图像尺寸变化敏感与处理图像位置信息方面却不如CNN,因此,近几年基于Transformer与CNN的网络结构在CV领域是研究热点。图像去雾技术可用于车载避让系统、遥感系统以及物联网设备被设为公共领域的热点问题,因此,本文提出基于Transformer的网络结构实现图像去雾。结合Transformer与UNet构建网络框架,针对Self-Attention因二次判别导致高计算复杂度的问题,我们使用聚焦线性注意力替换Softmax函数,将二次计算复杂度降为O(n),并且为了解决使用线性注意力引起性能下降的问题,使用恢复模块增加输出特征的多样性,实现性能的恢复。经实验表明,与现有的去雾霾网络相比,文章提出的方法在公开数据集RESIDE上获得了更好的实验结果以及更愉悦的去雾效果。

基于Transformer-LSTM网络的轴承寿命预测

轴承是旋转机械设备中的重要部件,由于工况、材质、加工方式等原因,轴承寿命时长相差许多。传统的并行或串行神经网络预测方式,对数据集有一定要求。因此,需要一种能够适用于不同数据长短的轴承剩余使用寿命预测网络。为此提出了一种能够预测不同寿命时长的Transformer-LSTM串并行神经网络预测模型。通过将Transformer解码层进行重构,并与长短期记忆时序神经网络(long short-term memory,LSTM)网络结构融合,实现轴承寿命数据的串并行预测处理。试验结果表明Transformer-LSTM神经网络能够精准预测长、中、短不同寿命时长的轴承失效时间,具有较强的模型泛化能力,提升轴承寿命预测精度与模型的泛化能力。

基于局部Transformer的泰语分词和词性标注联合模型

泰语分词和词性标注任务二者之间存在高关联性,已有研究表明将分词和词性标注任务进行联合学习可以有效提升模型性能,为此,提出了一种针对泰语拼写和构词特点的分词和词性标注联合模型。针对泰语中字符构成音节,音节组成词语的特点,采用局部Transformer网络从音节序列中学习分词特征;考虑到词根和词缀等音节与词性的关联,将用于分词的音节特征融入词语序列特征,缓解未知词的词性标注特征缺失问题。在此基础上,模型采用线性分类层预测分词标签,采用线性条件随机场建模词性序列的依赖关系。在泰语数据集LST20上的试验结果表明,模型分词F1、词性标注微平均F1和宏平均F1分别达到96.33%、97.06%和85.98%,相较基线模型分别提升了0.33%、0.44%和0.12%。

A Multi-Gap Multi-Channel Gas Switch for the Linear Transformer Driver

A multi-gap and multi-channel gas switch with convexo-convex discal planet electrodes was designed and investigated. Eight gaps are formed in series by a trigger electrode, six intermediate electrodes and two high voltage electrodes with a uniform gap length of 5 ram. The self breakdown and triggered breakdown performance of the switch are reported. Both the delay time and jitter decrease with the increase in the trigger voltage, switching coefficient and the decrease in the trigger isolating resistor. The delay time of the switch is about 40 ns, and the jitter is less than 2 ns when charged with 4-85 kV and triggered by a voltage pule of -75 kV. The inductance of the switch is about 30 nH.

Numerical Analysis of the Output-Pulse Shaping Capability of Linear Transformer Drivers

Output-pulse shaping capability of a linear transformer driver （LTD） module under different conditions is studied, by conducting the whole circuit model simulation by using the PSPICE code. Results indicate that a higher impedance profile of the internal transmission line would lead to a wider adjustment range for the output current rise time and a narrower adjustment range for the current peak. The number of cavities in series has a positive effect on the output- pulse shaping capability of LTD. Such an improvement in the output-pulse shaping capability can primarily be ascribed to the increment in the axial electric length of LTD. For a triggering time interval longer than the time taken by a pulse to propagate through the length of one cavity, the output parameters of LTD could be improved significantly. The present insulating capability of gas switches and other elements in the LTD cavities may only tolerate a slightly longer deviation in the triggering time interval. It is feasible for the LTD module to reduce the output current rise time, though it is not useful to improve the peak power effectively.

Influences of Switching Jitter on the Operational Performances of Linear Transformer Drivers-Based Drivers

A whole circuit model of a linear transformer drivers （LTD） module composed of 60 cavities in series was developed in the software PSPICE to study the influence of switching jitter on the operational performances of LTDs. In the model, each brick in each cavity is capable of operating with jitter in its switch. Additionally, the manner of triggering cables entering into cavities was considered. The performances of the LTD module operating with three typical cavity-triggering sequences were simulated and the simulation results indicate that switching jitter affects slightly the peak and starting time of the output current pulse. However, the enhancement in switching jitter would significantly lengthen the rise time of the output current pulse. Without considering other factors, a jitter lower than 10 ns may be necessary for the switches in the LTD module to provide output current parameters with an acceptable deviation.

Optical Diagnostics of Multi-Gap Gas Switches for Linear Transformer Drivers

The trigger characteristics of a multi-gap gas switch with double insulating layers,a square-groove electrode supporter and a UV pre-ionizing structure are investigated aided by a high sensitivity fiber-bundle array detector, a UV fiber detector, and a framing camera, in addition to standard electrical diagnostics. The fiber-bundle-array detector is used to track the turn-on sequence of each electrode gap at a timing precision of 0.6 ns. Each fiber bundle, including five fibers with different azimuth angles, aims at the whole emitting area of each electrode gap and is fed to a photomultiplier tube. The UV fiber detector with a spectrum response of 260-320 nm,including a fused-quartz fiber of 200 μm in diameter and a solar-blinded photomultiplier tube, is adopted to study the effect of UV pre-ionizing on trigger characteristics. The framing camera,with a capacity of 4 frames per shot and an exposure time of 5 ns, is employed to capture the evolution of channel arcs. Based on the turn-on light signal of each electrode gap, the breakdown delay is divided into statistical delay and formative delay. A decrease in both of them, a smaller switch jitter and more channel arcs are observed with lower gas pressure. An increase in trigger voltage can reduce the statistical delay and its jitter, while higher trigger voltage has a relatively small influence on the formative delay and the number of channel arcs. With the UV pre-ionizing structure at 0.24 MPa gas pressure and 60 kV trigger voltage, the statistical delay and its jitter can be reduced by 1.8 ns and 0.67 ns, while the formative delay and its jitter can only be reduced by 0.5 ns and 0.25 ns.

A Weakly-Supervised Crowd Density Estimation Method Based on Two-Stage Linear Feature Calibration

In a crowd density estimation dataset,the annotation of crowd locations is an extremely laborious task,and they are not taken into the evaluation metrics.In this paper,we aim to reduce the annotation cost of crowd datasets,and propose a crowd density estimation method based on weakly-supervised learning,in the absence of crowd position supervision information,which directly reduces the number of crowds by using the number of pedestrians in the image as the supervised information.For this purpose,we design a new training method,which exploits the correlation between global and local image features by incremental learning to train the network.Specifically,we design a parent-child network(PC-Net)focusing on the global and local image respectively,and propose a linear feature calibration structure to train the PC-Net simultaneously,and the child network learns feature transfer factors and feature bias weights,and uses the transfer factors and bias weights to linearly feature calibrate the features extracted from the Parent network,to improve the convergence of the network by using local features hidden in the crowd images.In addition,we use the pyramid vision transformer as the backbone of the PC-Net to extract crowd features at different levels,and design a global-local feature loss function(L2).We combine it with a crowd counting loss(LC)to enhance the sensitivity of the network to crowd features during the training process,which effectively improves the accuracy of crowd density estimation.The experimental results show that the PC-Net significantly reduces the gap between fullysupervised and weakly-supervised crowd density estimation,and outperforms the comparison methods on five datasets of Shanghai Tech Part A,ShanghaiTech Part B,UCF_CC_50,UCF_QNRF and JHU-CROWD++.

Experimental Investigation of the Deformation of Helical Gears with Double－Circular－Arc Tooth Profile Using Linear Variable Differential Transformer

ＥｘｐｅｒｉｍｅｎｔａｌＩｎｖｅｓｔｉｇａｔｉｏｎｏｆｔｈｅＤｅｆｏｒｍａｔｉｏｎｏｆＨｅｌｉｃａｌＧｅａｒｓｗｉｔｈＤｏｕｂｌｅ－Ｃｉｒｃｕｌａｒ－ＡｒｃＴｏｏｔｈＰｒｏｆｉｌｅＵｓｉｎｇＬｉｎｅａｒＶａｒｉａｂｌｅＤｉｆｆｅｒｅｎｔｉａｌＴｒａｎ...

Effect of New Suggested Ferroresonance Limiter on the Stability Domain of Chaotic Ferroresonance in the Power Transformer with Linear Core Model

This Paper studies the effect of new suggested ferroresonance limiter on controlling ferroresonance oscillations in the power transformer. It is expected that this limiter generally can control the ferroresonance. For studying these phenomena, at first ferroresonance is introduced and a general modeling approach is given. A simple case of ferroresonance in a three phase transformer is used to illustrate these phenomena. Then, effect of new suggested ferroresonance limiter on the onset of chaotic ferroresonance and control of these oscillations in a power transformer including linear core losses is studied. Simulation is done on a three phase power transformer while one of its phases is opened, and effect of varying input voltage on occurring ferroresonance overvoltage is studied. Results show that connecting the ferroresonance limiter to the transformer exhibits a great controlling effect on the ferroresonance overvoltage. Phase plane diagram, FFT analysis along with bifurcation diagrams are also presented. Significant effect on occurring chaotic ferroresonance, the range of parameter values that may lead to overvoltage and magnitude of ferroresonance overvoltage is obtained, showed and tabulated.

Transformer Characteristics of Linear Motor-Transformer Apparatus

The characteristics of linear transformer are studied analytically. The transformer is composed in one of modes of linear motor-transformer apparatus proposed for future wireless light rail vehicle (LRV). The secondary (onboard) power factor can be adjusted at any value by an onboard converter. The equivalent circuit is used to study the transferred power control. The parameters are determined by three-dimensional finite element method (FEM) analysis for one pole-pair model. Under the rated primary (input) and secondary voltage and current, which are specified for linear motor operation, the characteristics of the secondary power factor are cleared. It is also shown that the input capacitor can improve the primary power factor and decrease the input power capacity, but does not change the efficiency. This linear transformer has the efficiency of 91% and the input power factor of 0.87 when the apparatus without input capacitor is controlled at the secondary power factor of 0.4.

Driver Steering Behaviour Modelling Based on Neuromuscular Dynamics and Multi‑Task Time‑Series Transformer

Driver steering intention prediction provides an augmented solution to the design of an onboard collaboration mechanism between human driver and intelligent vehicle.In this study,a multi-task sequential learning framework is developed to pre-dict future steering torques and steering postures based on upper limb neuromuscular electromyography signals.The joint representation learning for driving postures and steering intention provides an in-depth understanding and accurate modelling of driving steering behaviours.Regarding different testing scenarios,two driving modes,namely,both-hand and single-right-hand modes,are studied.For each driving mode,three different driving postures are further evaluated.Next,a multi-task time-series transformer network(MTS-Trans)is developed to predict the future steering torques and driving postures based on the multi-variate sequential input and the self-attention mechanism.To evaluate the multi-task learning performance and information-sharing characteristics within the network,four distinct two-branch network architectures are evaluated.Empirical validation is conducted through a driving simulator-based experiment,encompassing 21 participants.The pro-posed model achieves accurate prediction results on future steering torque prediction as well as driving posture recognition for both two-hand and single-hand driving modes.These findings hold significant promise for the advancement of driver steering assistance systems,fostering mutual comprehension and synergy between human drivers and intelligent vehicles.

An Improved Robust Model Predictive Control Approach to Systems with Linear Fractional Transformation Perturbations

In this paper, a robust model predictive control approach is proposed for a class of uncertain systems with time-varying, linear fractional transformation perturbations. By adopting a sequence of feedback control laws instead of a single one, the control performance can be improved and the region of attraction can be enlarged compared with the existing model predictive control （MPC） approaches. Moreover, a synthesis approach of MPC is developed to achieve high performance with lower on-line computational burden. The effectiveness of the proposed approach is verified by simulation examples.

Linear Complementary Method for Moving Boundary Problems with Phase Transformation

In this paper, the linear complementary method for moving boundary problems with phase transformation is presented, in which a pair of unknown vectors of heat source with phase transforming and the temperature field can be solved exactly, and a large amount of iterative calculations can be avoided.

Uncertainty Principle for the Quaternion Linear Canonical Transform in Terms of Covariance

An uncertainty principle(UP),which offers information about a signal and its Fourier transform in the time-frequency plane,is particularly powerful in mathematics,physics and signal processing community.Under the polar coordinate form of quaternion-valued signals,the UP of the two-sided quaternion linear canonical transform(QLCT)is strengthened in terms of covariance.The condition giving rise to the equal relation of the derived result is obtained as well.The novel UP with covariance can be regarded as one in a tighter form related to the QLCT.It states that the product of spreads of a quaternion-valued signal in the spatial domain and the QLCT domain is bounded by a larger lower bound.

Transformer网络在雷达辐射源识别中的应用

由于雷达辐射源信号在时序上具有强相关性,针对卷积神经网络难以提取信号的时序逻辑相关性特征的缺点,本文提出了基于Transformer网络的雷达辐射源个体识别的方法。该方法首先以3台信号发生器模拟雷达辐射源信号,同时采集数据并建立数据库;然后对原始信号提取相位特征以及包络特征;最后采用基于注意力机制的Transformer网络对信号进行进一步的特征提取并分类。实验结果表明,该算法对于包络特征以及相位特征敏感度不同,同时数据样本长度对于识别效果会产生影响,最终结合相位特征的网络模型在信噪比为15 dB时识别率可以达到98.9%。

CONSTRUCTION OF PROXY BLIND SIGNATURE SCHEME BASED ON MULTI-LINEAR TRANSFORM

A general method of constructing proxy blind signature is proposed based on multilinear transform. Based on this method, the four proxy blind signature schemes are correspondently generated with four different signature equations, and each of them has four forms of variations of signs. Hence there are sixteen signatures in all, and all of them are proxy stronglyblind signature schemes. Furthermore, the two degenerated situations of multi-linear transform are discussed. Their corresponding proxy blind signature schemes are shown, too. But some schemes come from one of these degenerate situations are proxy weakly-blind signature scheme.The security for proposed scheme is analyzed in details. The results indicate that these signature schemes have many good properties such as unforgeability, distinguish-ability of proxy signature,non-repudiation and extensive value of application etc.

THE BLUE AND MINQUE IN GAUSS-MARKOFF MODEL WITH LINEAR TRANSFORMATION OF THE OBSERVABLE VARIABLES

For a singular linear model A = （y, Xβ, σ2 V） and its transformed model MF = （Fy, FXβ, σ 2FVF＇）, where V is nonnegative definite and X can be rank-deficient, the expressions for the differences of the estimates for the vector of FXβ and the variance factor σ2 are given. Moreover, the necessary and sufficient conditions for the equalities of the estimates for the vector of FXβ and the variance factor σ2 are also established. In the meantime, works in Baksalary and Kala （1981） are strengthened and consequences in Puntanen and Nurhonen （1992）, and Puntanen （1996） are extended.

Predicting Nonlinear Wave Trough Distributions Utilizing A Transformed Linear Simulation Method

This paper first proposes a new approach for predicting the nonlinear wave trough distributions by utilizing a transformed linear simulation method. The linear simulation method is transformed based on a Hermite transformation model where the transformation is chosen to be a monotonic cubic polynomial and calibrated such that the first four moments of the transformed model match the moments of the true process. The proposed new approach is applied for calculating the wave trough distributions of a nonlinear sea state with the surface elevation data measured at the coast of Yura in the Japan Sea, and its accuracy and efficiency are convincingly validated by comparisons with the results from two theoretical distribution models, from a linear simulation model and a secondorder nonlinear simulation model. Finally, it is further demonstrated in this paper that the new approach can be applied to all the situations characterized by similar nondimensional spectrum.

NOVEL LINEAR TRANSFORMATION SWITCHED-CAPACITOR FILTER DESIGN

This paper presents a novel method that is applied to realize the Linear Transformation(LT)Switched-Capacitor Filter(SCF).It adopts the Voltage Control Voltage Source(VCVS)equalized transfor-mation to revise the original LC ladder filter and induce it into 16 basic sections and then extend the princi-ple of the LT in order to fit active and 3 port networks and give out switched-capacitor circuits corre-sponding to the 16 basic sections,which can realize all four kinds of filters——LP,HP,BP,BS filters.De-signed examples are given here.An Nth order filter only requires N amplifiers and the circuit is insensitive toparasitic capacitances.The experimental results of a 3rd order elliptic LP and a 6th order elliptic BP are giv-en and agree with the theory.