目前自闭症功能磁共振(functional magnetic resonance imaging,fMRI)图像分类模型在跨多个机构的数据集下分类精度较低,难以应用到自闭症的诊断工作中。为此,本文提出了一种基于Transformer的自闭症分类模型(autism spectrum disorder ...目前自闭症功能磁共振(functional magnetic resonance imaging,fMRI)图像分类模型在跨多个机构的数据集下分类精度较低,难以应用到自闭症的诊断工作中。为此,本文提出了一种基于Transformer的自闭症分类模型(autism spectrum disorder classification model based on Transformer,TransASD)。首先采用脑图谱模板提取fMRI数据中的时间序列输入Transformer模型,并引入一种重叠窗口注意力机制,能够更好地捕捉异构数据的局部与全局特征。其次,提出了一个跨窗口正则化方法作为额外的损失项,使模型可以更加准确地聚焦于重要的特征。本文使用该模型在公开的自闭症数据集ABIDE上进行实验,在10折交叉验证法下得到了71.44%的准确率,该模型对比其他先进算法模型取得了更好的分类效果。展开更多
针对多变量时序(Multivariate Time Series,MTS)分类中长序列数据难以捕捉时序特征的问题,提出一种基于双向稀疏Transformer的时序分类模型BST(Bidirectional Sparse Transformer),提高了MTS分类任务的准确度.BST模型使用Transformer框...针对多变量时序(Multivariate Time Series,MTS)分类中长序列数据难以捕捉时序特征的问题,提出一种基于双向稀疏Transformer的时序分类模型BST(Bidirectional Sparse Transformer),提高了MTS分类任务的准确度.BST模型使用Transformer框架,构建了一种基于活跃度得分的双向稀疏注意力机制.基于KL散度构建活跃度评价函数,并将评价函数的非对称问题转变为对称权重问题.据此,对原有查询矩阵、键值矩阵进行双向稀疏化,从而降低原Transformer模型中自注意力机制运算的时间复杂度.实验结果显示,BST模型在9个长序列数据集上取得最高平均排名,在临界差异图中领先第2名35.7%,对于具有强时序性的乙醇浓度数据集(Ethanol Concentration,EC),分类准确率提高30.9%.展开更多
In order to provide larger capacity of the hidden secret data while maintaining a good visual quality of stego-image, in accordance with the visual property that human eyes are less sensitive to strong texture, a nove...In order to provide larger capacity of the hidden secret data while maintaining a good visual quality of stego-image, in accordance with the visual property that human eyes are less sensitive to strong texture, a novel steganographic method based on wavelet and modulus function is presented. First, an image is divided into blocks of prescribed size, and every block is decomposed into one-level wavelet. Then, the capacity of the hidden secret data is decided with the number of wavelet coefficients of larger magnitude. Finally, secret information is embedded by steganography based on modulus function. From the experimental results, the proposed method hides much more information and maintains a good visual quality of stego-image. Besides, the embedded data can be extracted from the stego-image without referencing the original image.展开更多
Power transformer insulation systems are subjected to many stresses during normal operation due to lightning and switching.If the spectrum of incoming surge voltage matches the winding one,the corresponding resonance ...Power transformer insulation systems are subjected to many stresses during normal operation due to lightning and switching.If the spectrum of incoming surge voltage matches the winding one,the corresponding resonance will be excited.Therefore external transients occurring in power systems might trigger internal overvoltages with large maximum value in transformer windings.Overvoltages having such characteristic have been the root cause of many power transformer failures.The paper presents an approach to the identification of sensitive zones in the transformer windings based on the measurements of overvoltages inside windings and frequency dependences of admittance of the power transformer.The frequency characteristic of the transformer winding may determine those regions in the frequency spectrum.The presented approach might be used both for design optimization and diagnostics of distribution and power transformers.展开更多
由于水下环境的多样性和光在水中受到的散射及选择性吸收作用,采集到的水下图像通常会产生严重的质量退化问题,如颜色偏差、清晰度低和亮度低等,为解决以上问题,本文提出了一种基于Transformer和生成对抗网络的水下图像增强算法。以生...由于水下环境的多样性和光在水中受到的散射及选择性吸收作用,采集到的水下图像通常会产生严重的质量退化问题,如颜色偏差、清晰度低和亮度低等,为解决以上问题,本文提出了一种基于Transformer和生成对抗网络的水下图像增强算法。以生成对抗网络为基础架构,结合编码解码结构、基于空间自注意力机制的全局特征建模Transformer模块和通道级多尺度特征融合Transformer模块构建了TGAN(generative adversarial network with transformer)网络增强模型,重点关注水下图像衰减更严重的颜色通道和空间区域,有效增强了图像细节并解决了颜色偏差问题。此外,设计了一种结合RGB和LAB颜色空间的多项损失函数,约束网络增强模型的对抗训练。实验结果表明,与CLAHE(contrast limited adaptive histogram equalization)、UDCP(underwater dark channel prior)、UWCNN(underwater based on convolutional neural network)、FUnIE-GAN(fast underwater image enhancement for improved visual perception)等典型水下图像增强算法相比,所提算法增强后的水下图像在清晰度、细节纹理和色彩表现等方面都有所提升,客观评价指标如峰值信噪比、结构相似性和水下图像质量度量的平均值分别提升了5.8%、1.8%和3.6%,有效地提升了水下图像的视觉感知效果。展开更多
Fractional Fourier transform(FRFT)is a linear transform generalizing Fourier transform(FT)that plays an important role in the field of signal processing and analysis.FRFT contains an adjustable parameterα,which it ro...Fractional Fourier transform(FRFT)is a linear transform generalizing Fourier transform(FT)that plays an important role in the field of signal processing and analysis.FRFT contains an adjustable parameterα,which it rotates the signal in the time frequency plane and represents the signal in an intermediate domain between time and frequency.FRFT provides a measure about the angular distribution of signal’s energy in time frequency plane.FT is a special case of FRFT when angleαis equal toπ/2.This paper presents mathematical model for obtaining FRFT of PC6 window function.The different parameters of this window function are also obtained with the help of simulation results.A comparison of window function parameters is presented using FT and FRFT.Also comparison of this window function with Hanning window function is presented in terms of Side Lobe Fall off Rate(SLFOR).For different values of FRFT order,PC6 window function shows variation in different parameters.Thus by changing the FRFT order,the minimum stop band attenuation of the resulting window function can be controlled.展开更多
文章以幂函数变换为研究对象,从背景值误差和还原误差的角度分析了幂函数变换对GM(1,1)模型建模精度的影响,论证了幂函数变换的GM(1,1)模型(PFNGM(1,1)模型)具有逼近无偏性,能在可忽略的误差范围内实现对白指数序列的预测无偏性。实例...文章以幂函数变换为研究对象,从背景值误差和还原误差的角度分析了幂函数变换对GM(1,1)模型建模精度的影响,论证了幂函数变换的GM(1,1)模型(PFNGM(1,1)模型)具有逼近无偏性,能在可忽略的误差范围内实现对白指数序列的预测无偏性。实例应用结果表明,其建模精度和预测效果均优于无偏GM(1,1)模型和离散GM(1,1)模型。为将适宜建模序列拓展至近似非齐次指数序列和季节波动序列,同时保留幂函数变换可以有效降低背景值误差对建模精度影响的优势,将幂函数变换与平移变换相结合构建了PFNGM(1,1)模型,将幂函数变换与季节性GM(1,1)模型(SGM(1,1)模型)相结合构建了PFSGM(1,1)模型。实例应用结果表明,PFNGM(1,1)模型的建模精度和预测效果均优于背景值改进的NGM(1,1, k )模型和ONGM(1,1, k,c )模型,PFSGM(1,1)模型的建模精度和预测效果均优于SGM(1,1)模型,验证了两种模型的有效性。展开更多
文摘目前自闭症功能磁共振(functional magnetic resonance imaging,fMRI)图像分类模型在跨多个机构的数据集下分类精度较低,难以应用到自闭症的诊断工作中。为此,本文提出了一种基于Transformer的自闭症分类模型(autism spectrum disorder classification model based on Transformer,TransASD)。首先采用脑图谱模板提取fMRI数据中的时间序列输入Transformer模型,并引入一种重叠窗口注意力机制,能够更好地捕捉异构数据的局部与全局特征。其次,提出了一个跨窗口正则化方法作为额外的损失项,使模型可以更加准确地聚焦于重要的特征。本文使用该模型在公开的自闭症数据集ABIDE上进行实验,在10折交叉验证法下得到了71.44%的准确率,该模型对比其他先进算法模型取得了更好的分类效果。
基金the National Natural Science Foundation of China (50677014)Hunan Provincial Natural Science Foundation of China (06JJ50114).
文摘In order to provide larger capacity of the hidden secret data while maintaining a good visual quality of stego-image, in accordance with the visual property that human eyes are less sensitive to strong texture, a novel steganographic method based on wavelet and modulus function is presented. First, an image is divided into blocks of prescribed size, and every block is decomposed into one-level wavelet. Then, the capacity of the hidden secret data is decided with the number of wavelet coefficients of larger magnitude. Finally, secret information is embedded by steganography based on modulus function. From the experimental results, the proposed method hides much more information and maintains a good visual quality of stego-image. Besides, the embedded data can be extracted from the stego-image without referencing the original image.
文摘Power transformer insulation systems are subjected to many stresses during normal operation due to lightning and switching.If the spectrum of incoming surge voltage matches the winding one,the corresponding resonance will be excited.Therefore external transients occurring in power systems might trigger internal overvoltages with large maximum value in transformer windings.Overvoltages having such characteristic have been the root cause of many power transformer failures.The paper presents an approach to the identification of sensitive zones in the transformer windings based on the measurements of overvoltages inside windings and frequency dependences of admittance of the power transformer.The frequency characteristic of the transformer winding may determine those regions in the frequency spectrum.The presented approach might be used both for design optimization and diagnostics of distribution and power transformers.
文摘由于水下环境的多样性和光在水中受到的散射及选择性吸收作用,采集到的水下图像通常会产生严重的质量退化问题,如颜色偏差、清晰度低和亮度低等,为解决以上问题,本文提出了一种基于Transformer和生成对抗网络的水下图像增强算法。以生成对抗网络为基础架构,结合编码解码结构、基于空间自注意力机制的全局特征建模Transformer模块和通道级多尺度特征融合Transformer模块构建了TGAN(generative adversarial network with transformer)网络增强模型,重点关注水下图像衰减更严重的颜色通道和空间区域,有效增强了图像细节并解决了颜色偏差问题。此外,设计了一种结合RGB和LAB颜色空间的多项损失函数,约束网络增强模型的对抗训练。实验结果表明,与CLAHE(contrast limited adaptive histogram equalization)、UDCP(underwater dark channel prior)、UWCNN(underwater based on convolutional neural network)、FUnIE-GAN(fast underwater image enhancement for improved visual perception)等典型水下图像增强算法相比,所提算法增强后的水下图像在清晰度、细节纹理和色彩表现等方面都有所提升,客观评价指标如峰值信噪比、结构相似性和水下图像质量度量的平均值分别提升了5.8%、1.8%和3.6%,有效地提升了水下图像的视觉感知效果。
文摘Fractional Fourier transform(FRFT)is a linear transform generalizing Fourier transform(FT)that plays an important role in the field of signal processing and analysis.FRFT contains an adjustable parameterα,which it rotates the signal in the time frequency plane and represents the signal in an intermediate domain between time and frequency.FRFT provides a measure about the angular distribution of signal’s energy in time frequency plane.FT is a special case of FRFT when angleαis equal toπ/2.This paper presents mathematical model for obtaining FRFT of PC6 window function.The different parameters of this window function are also obtained with the help of simulation results.A comparison of window function parameters is presented using FT and FRFT.Also comparison of this window function with Hanning window function is presented in terms of Side Lobe Fall off Rate(SLFOR).For different values of FRFT order,PC6 window function shows variation in different parameters.Thus by changing the FRFT order,the minimum stop band attenuation of the resulting window function can be controlled.
文摘文章以幂函数变换为研究对象,从背景值误差和还原误差的角度分析了幂函数变换对GM(1,1)模型建模精度的影响,论证了幂函数变换的GM(1,1)模型(PFNGM(1,1)模型)具有逼近无偏性,能在可忽略的误差范围内实现对白指数序列的预测无偏性。实例应用结果表明,其建模精度和预测效果均优于无偏GM(1,1)模型和离散GM(1,1)模型。为将适宜建模序列拓展至近似非齐次指数序列和季节波动序列,同时保留幂函数变换可以有效降低背景值误差对建模精度影响的优势,将幂函数变换与平移变换相结合构建了PFNGM(1,1)模型,将幂函数变换与季节性GM(1,1)模型(SGM(1,1)模型)相结合构建了PFSGM(1,1)模型。实例应用结果表明,PFNGM(1,1)模型的建模精度和预测效果均优于背景值改进的NGM(1,1, k )模型和ONGM(1,1, k,c )模型,PFSGM(1,1)模型的建模精度和预测效果均优于SGM(1,1)模型,验证了两种模型的有效性。
