高光谱成像技术的飞速发展给非侵入式医学成像带来新的契机,但高光谱医学图像具有高维度、高冗余以及“图谱合一”的特点,亟需针对上述特点设计智能诊断算法。近年来,Transformer已经在高光谱医学图像处理领域得到广泛应用。然而,不同...高光谱成像技术的飞速发展给非侵入式医学成像带来新的契机,但高光谱医学图像具有高维度、高冗余以及“图谱合一”的特点,亟需针对上述特点设计智能诊断算法。近年来,Transformer已经在高光谱医学图像处理领域得到广泛应用。然而,不同仪器设备、不同采集操作所获得的高光谱医学图像差异较大,这给现有Transformer诊断模型的实际应用带来了巨大挑战。针对上述问题,本文提出了一种空-谱自注意力Transformer(S3AT),自适应挖掘像素与像素间、波段与波段间的内蕴联系,并在分类阶段融合多个视野下的预测结果。首先,在Transformer编码器中,设计一种空-谱自注意力机制,获取不同视野下高光谱图像上的关键空间信息和重要波段,并将不同视野下所获得的空-谱自注意力进行融合。其次,在模型分类阶段,将不同视野下的预测结果根据可学习权重进行加权融合,对图像进行综合预测。在In-vivo Human Brain和BloodCell HSI两个数据集上,本文算法总体分类精度分别达到82.25%和91.74%。实验结果表明,所提出的算法有效改善高光谱医学图像分类性能。展开更多
负荷预测是综合能源系统(integrated energy system,IES)能量管理和优化调度的基础,其预测精度直接关系到系统的整体运行性能。提出了一种基于Transformer网络和多任务学习的园区综合能源系统电-热短期负荷预测模型。首先对Transformer...负荷预测是综合能源系统(integrated energy system,IES)能量管理和优化调度的基础,其预测精度直接关系到系统的整体运行性能。提出了一种基于Transformer网络和多任务学习的园区综合能源系统电-热短期负荷预测模型。首先对Transformer网络和多任务学习结构的基本原理进行了介绍;然后通过基于随机森林的特征选择步骤提取反映负荷特性和变化规律的典型指标,构建多任务学习输入特征,基于Transformer网络构建多任务学习权值共享层,并通过全连接层输出多能负荷的预测值;最后通过实际园区微能源系统的数据验证所提方法和算法的有效性,结果表明本文所提模型可以充分学习电-热耦合特征,提高负荷预测的精度。展开更多
An effective near-field - far-field (NF - FF) transformation with spherical scanning for quasi-planar antennas from irregularly spaced data is developed in this paper. Two efficient approaches for evaluating the regul...An effective near-field - far-field (NF - FF) transformation with spherical scanning for quasi-planar antennas from irregularly spaced data is developed in this paper. Two efficient approaches for evaluating the regularly spaced spherical samples from the nonuniformly distributed ones are proposed and numerically compared. Both the approaches rely on a nonredundant sampling representation of the voltage measured by the probe, based on an oblate ellipsoidal modelling of the antenna under test. The former employs the singular value decomposition method to reconstruct the NF data at the points fixed by the nonredundant sampling representation and can be applied when the irregularly acquired samples lie on nonuniform parallels. The latter is based on an iterative technique and can be used also when such a hypothesis does not hold, but requires the existence of a biunique correspondence between the uniform and nonuniform samples, associ- ating at each uniform sampling point the nearest irregular one. Once the regularly spaced spherical samples have been recovered, the NF data needed by a probe compensated NF - FF transformation with spherical scanning are efficiently evaluated by using an optimal sampling interpolation algorithm. It is so possible to accurately compensate known posi- tioning errors in the NF - FF transformation with spherical scanning for quasi-planar antennas. Some numerical tests assessing the accuracy and the robustness of the proposed approaches are reported.展开更多
In recent years, Empirical mode decomposition and Hilbert spectral analysis have been combined to identify system parameters. Singular-Value Decomposition is pro- posed as a signal preprocessing technique of Hilbert-H...In recent years, Empirical mode decomposition and Hilbert spectral analysis have been combined to identify system parameters. Singular-Value Decomposition is pro- posed as a signal preprocessing technique of Hilbert-Huang Transform to extract modal parameters for closely spaced modes and low-energy components. The proposed method is applied to a simulated airplane model built in Automatic Dynamic Analysis of Mechanical Systems software. The results demonstrate that the identified modal parameters are in good agreement with the baseline model.展开更多
By means of riM-190 hot-stage microscopy,the in situ observation of α-β transformation in 12Cr2MoWVTiB steel has been carried out.The sequence of the growth of ferritic needles composing bainitic basket has also bee...By means of riM-190 hot-stage microscopy,the in situ observation of α-β transformation in 12Cr2MoWVTiB steel has been carried out.The sequence of the growth of ferritic needles composing bainitic basket has also been determined.According to the crystallographic analy- sis a multislip system transform model concerning the formation of basket has been proposed.展开更多
细粒度图像分类是计算机视觉领域的一大分类任务,其难点在于如何通过类别监督信息自主地找到判别性区域.提出一种新的通道-空间融合注意力模块,基于该模块设计了一种新的Swin Transformer算法SwinT⁃NCSA(a Swin Transformer based on a ...细粒度图像分类是计算机视觉领域的一大分类任务,其难点在于如何通过类别监督信息自主地找到判别性区域.提出一种新的通道-空间融合注意力模块,基于该模块设计了一种新的Swin Transformer算法SwinT⁃NCSA(a Swin Transformer based on a novel channel⁃spatial attention module),分别从通道维和空间维同时提取特征,再将其融入到Swin Transformer模型中以提高其小尺度中多头注意力信息的提取能力.SwinT⁃NCSA算法特别关注了对分类有用的区域,同时忽视对分类无用的背景区域,以此在细粒度图像分类任务中达到较高的分类准确率.在FGVC Aircraft飞机数据集、CUB-200-2011鸟类数据集和Stanford Cars车类数据集3个公共数据集上的实验表明,SwinT⁃NCSA算法可以分别取得93.3%、88.4%和94.7%的准确率,优于同类算法.展开更多
Under the Flaschka-Newell Lax pair,the Darboux transformation for the Painlevé-Ⅱequation is constructed by the limiting technique.With the aid of the Darboux transformation,the rational solutions are represented...Under the Flaschka-Newell Lax pair,the Darboux transformation for the Painlevé-Ⅱequation is constructed by the limiting technique.With the aid of the Darboux transformation,the rational solutions are represented by the Gram determinant,and then we give the large y asymptotics of the determinant and the rational solutions.Finally,the solution of the corresponding Riemann-Hilbert problem is obtained from the Darboux matrices.展开更多
为了提高光伏发电功率预测精度,提出了一种基于长短期时序数据融合的Transformer生成式预测模型:LSTformer,能准确有效地预测光伏发电功率。LSTformer创新性地提出了时序分析模块(time series analysis,TSA)、时序特征融合模块(time ser...为了提高光伏发电功率预测精度,提出了一种基于长短期时序数据融合的Transformer生成式预测模型:LSTformer,能准确有效地预测光伏发电功率。LSTformer创新性地提出了时序分析模块(time series analysis,TSA)、时序特征融合模块(time series feature fusion,TSFF)和多周期嵌入模块(cycleEmbed),利用数据融合解决难以提取多时间尺度时序特征问题。设计时间卷积前馈(time convolution feedforward,TCNforward)单元,在编解码的过程中进一步提取时序特征。利用某光伏电站实际历史发电数据,通过实验验证LSTformer模型在光伏发电功率预测领域得到最低的均方误差(mean squared error,MSE)、平均绝对误差(mean absolute error,MAE),并通过消融实验验证了各模块的有效性。展开更多
文摘高光谱成像技术的飞速发展给非侵入式医学成像带来新的契机,但高光谱医学图像具有高维度、高冗余以及“图谱合一”的特点,亟需针对上述特点设计智能诊断算法。近年来,Transformer已经在高光谱医学图像处理领域得到广泛应用。然而,不同仪器设备、不同采集操作所获得的高光谱医学图像差异较大,这给现有Transformer诊断模型的实际应用带来了巨大挑战。针对上述问题,本文提出了一种空-谱自注意力Transformer(S3AT),自适应挖掘像素与像素间、波段与波段间的内蕴联系,并在分类阶段融合多个视野下的预测结果。首先,在Transformer编码器中,设计一种空-谱自注意力机制,获取不同视野下高光谱图像上的关键空间信息和重要波段,并将不同视野下所获得的空-谱自注意力进行融合。其次,在模型分类阶段,将不同视野下的预测结果根据可学习权重进行加权融合,对图像进行综合预测。在In-vivo Human Brain和BloodCell HSI两个数据集上,本文算法总体分类精度分别达到82.25%和91.74%。实验结果表明,所提出的算法有效改善高光谱医学图像分类性能。
文摘负荷预测是综合能源系统(integrated energy system,IES)能量管理和优化调度的基础,其预测精度直接关系到系统的整体运行性能。提出了一种基于Transformer网络和多任务学习的园区综合能源系统电-热短期负荷预测模型。首先对Transformer网络和多任务学习结构的基本原理进行了介绍;然后通过基于随机森林的特征选择步骤提取反映负荷特性和变化规律的典型指标,构建多任务学习输入特征,基于Transformer网络构建多任务学习权值共享层,并通过全连接层输出多能负荷的预测值;最后通过实际园区微能源系统的数据验证所提方法和算法的有效性,结果表明本文所提模型可以充分学习电-热耦合特征,提高负荷预测的精度。
文摘An effective near-field - far-field (NF - FF) transformation with spherical scanning for quasi-planar antennas from irregularly spaced data is developed in this paper. Two efficient approaches for evaluating the regularly spaced spherical samples from the nonuniformly distributed ones are proposed and numerically compared. Both the approaches rely on a nonredundant sampling representation of the voltage measured by the probe, based on an oblate ellipsoidal modelling of the antenna under test. The former employs the singular value decomposition method to reconstruct the NF data at the points fixed by the nonredundant sampling representation and can be applied when the irregularly acquired samples lie on nonuniform parallels. The latter is based on an iterative technique and can be used also when such a hypothesis does not hold, but requires the existence of a biunique correspondence between the uniform and nonuniform samples, associ- ating at each uniform sampling point the nearest irregular one. Once the regularly spaced spherical samples have been recovered, the NF data needed by a probe compensated NF - FF transformation with spherical scanning are efficiently evaluated by using an optimal sampling interpolation algorithm. It is so possible to accurately compensate known posi- tioning errors in the NF - FF transformation with spherical scanning for quasi-planar antennas. Some numerical tests assessing the accuracy and the robustness of the proposed approaches are reported.
文摘In recent years, Empirical mode decomposition and Hilbert spectral analysis have been combined to identify system parameters. Singular-Value Decomposition is pro- posed as a signal preprocessing technique of Hilbert-Huang Transform to extract modal parameters for closely spaced modes and low-energy components. The proposed method is applied to a simulated airplane model built in Automatic Dynamic Analysis of Mechanical Systems software. The results demonstrate that the identified modal parameters are in good agreement with the baseline model.
文摘By means of riM-190 hot-stage microscopy,the in situ observation of α-β transformation in 12Cr2MoWVTiB steel has been carried out.The sequence of the growth of ferritic needles composing bainitic basket has also been determined.According to the crystallographic analy- sis a multislip system transform model concerning the formation of basket has been proposed.
文摘细粒度图像分类是计算机视觉领域的一大分类任务,其难点在于如何通过类别监督信息自主地找到判别性区域.提出一种新的通道-空间融合注意力模块,基于该模块设计了一种新的Swin Transformer算法SwinT⁃NCSA(a Swin Transformer based on a novel channel⁃spatial attention module),分别从通道维和空间维同时提取特征,再将其融入到Swin Transformer模型中以提高其小尺度中多头注意力信息的提取能力.SwinT⁃NCSA算法特别关注了对分类有用的区域,同时忽视对分类无用的背景区域,以此在细粒度图像分类任务中达到较高的分类准确率.在FGVC Aircraft飞机数据集、CUB-200-2011鸟类数据集和Stanford Cars车类数据集3个公共数据集上的实验表明,SwinT⁃NCSA算法可以分别取得93.3%、88.4%和94.7%的准确率,优于同类算法.
基金Project supported by the National Natural Science Foundation of China (Grant No.12101246)。
文摘Under the Flaschka-Newell Lax pair,the Darboux transformation for the Painlevé-Ⅱequation is constructed by the limiting technique.With the aid of the Darboux transformation,the rational solutions are represented by the Gram determinant,and then we give the large y asymptotics of the determinant and the rational solutions.Finally,the solution of the corresponding Riemann-Hilbert problem is obtained from the Darboux matrices.
