Si-PIN photodetectors having features such as low cost,small size,low weight,low voltage,and low power consumption are widely used as radiation detectors in electronic personal dosimeters(EPDs).The technical parameter...Si-PIN photodetectors having features such as low cost,small size,low weight,low voltage,and low power consumption are widely used as radiation detectors in electronic personal dosimeters(EPDs).The technical parameters of EPDs based on the Si-PIN photodetectors include photon energy response(PER),angular response,inherent error,and dose rate linearity.Among them,PER is a key parameter for evaluation of EPD measurement accuracy.At present,owing to the limitations of volume,power consumption,and EPD cost,the PER is usually corrected by a combination of single-channel counting techniques and filtering material methods.However,the above-mentioned methods have problems such as poor PER and low measurement accuracy.To solve such problems,in this study,a 1024-channel spectrometry system using a Si-PIN photodetector was developed and fullspectrum measurement in the reference radiation fields was conducted for radiation protection.The measurement results using the few-channel spectroscopy dose method showed that the PER could be controlled within±14%and±2%under the conditions of two and three energy intervals,respectively,with different channel numbers.The PER measured at 0°angle of radiation incidence meets the-29%to+67%requirements of IEC 61526:2010.Meanwhile,the channel number and counts-to-dose conversion factors formed in the experiment can be integrated into an EPD.展开更多
在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效...在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效融合通道信息和空间信息,提升小样本细粒度图像分类的性能.通道先验感知模块实现了通道维度上注意力权重的动态分配,从而高效地捕捉通道先验信息;多尺度特征聚合过程充分利用细粒度图像中丰富的上下文信息,获取丰富的空间和边界细节特征;最后,特征细化模块对上述提取的通道和空间维度信息进行优化,实现了对关键区域的动态选择和强调,进而融合形成更精细、更具代表性的混合特征表示.所提算法在以Conv-4作为骨干网络时,在Stanford Cars、Stanford Dogs和CUB-200-2011三个细粒度数据集上的实验分类性能显著提升.在5 way 1 shot分类任务中,三个数据集的准确率分别达到了79.95%、66.97%和81.91%;在5 way 5 shot分类任务中,准确率则分别为93.42%、82.48%和93.19%.展开更多
在小样本学习任务中,针对传统的骨干卷积网络在提取图像特征时,由于多层卷积忽视细节特征导致特征信息丢失,因而图像分类准确率不高的问题,提出了基于两阶段特征空间增强的小样本图像分类模型。首先,该模型在残差网络(residual network,...在小样本学习任务中,针对传统的骨干卷积网络在提取图像特征时,由于多层卷积忽视细节特征导致特征信息丢失,因而图像分类准确率不高的问题,提出了基于两阶段特征空间增强的小样本图像分类模型。首先,该模型在残差网络(residual network, ResNet)12的底层引入中值增强的空间和通道注意力块(median-enhanced spatial and channel attention block, MESC);然后,该模型在ResNet12的中高层引入空间组增强(spatial group-wise enhance, SGE)模块,提升卷积神经网络中的语义特征学习能力,使模型有效提取特征图关键信息。该模型通过增强有限的训练样本的特征表示来提高分类性能,增强模型对噪声的鲁棒性。结果表明,该模型在加州理工学院-加利福尼亚大学圣地亚哥分校鸟类(California Institute of Technology-University of California at San Diego birds, CUB)-200-2011数据集上,5类别1样本和5类别5样本2种参数设置下的分类准确率分别比分布传播图网络(distribution propagation graph network, DPGN)模型提高了约5.15%和1.92%;在分层图像网络(tiered ImageNet, tieredImageNet)数据集上,这2种参数设置下的分类准确率分别比DPGN模型提高了约1.04%和0.55%。该模型提升了小样本图像分类任务的性能。展开更多
Background Electroencephalography(EEG)has gained popularity in various types of biomedical applications as a signal source that can be easily acquired and conveniently analyzed.However,owing to a complex scalp electri...Background Electroencephalography(EEG)has gained popularity in various types of biomedical applications as a signal source that can be easily acquired and conveniently analyzed.However,owing to a complex scalp electrical environment,EEG is often polluted by diverse artifacts,with electromyography artifacts being the most difficult to remove.In particular,for ambulatory EEG devices with a restricted number of channels,dealing with muscle artifacts is a challenge.Methods In this study,we propose a simple but effective novel scheme that combines singular spectrum analysis(SSA)and canonical correlation analysis(CCA)algorithms for single-channel problems and then extend it to a few channel case by adding additional combining and dividing operations to channels.Results We evaluated our proposed framework on both semi-simulated and real-life data and compared it with some state-of-the art methods.The results demonstrate this novel framework's superior performance in both single-channel and few-channel cases.Conclusions This promising approach,based on its effectiveness and low time cost,is suitable for real-world biomedical signal processing applications.展开更多
基金This work was partly supported by the National Key Scientific Instruments to Develop Dedicated Program(Nos.2013YQ090811 and 2016YFF0103800)the National Key Research and Development Program(No.2017YFF0211100).
文摘Si-PIN photodetectors having features such as low cost,small size,low weight,low voltage,and low power consumption are widely used as radiation detectors in electronic personal dosimeters(EPDs).The technical parameters of EPDs based on the Si-PIN photodetectors include photon energy response(PER),angular response,inherent error,and dose rate linearity.Among them,PER is a key parameter for evaluation of EPD measurement accuracy.At present,owing to the limitations of volume,power consumption,and EPD cost,the PER is usually corrected by a combination of single-channel counting techniques and filtering material methods.However,the above-mentioned methods have problems such as poor PER and low measurement accuracy.To solve such problems,in this study,a 1024-channel spectrometry system using a Si-PIN photodetector was developed and fullspectrum measurement in the reference radiation fields was conducted for radiation protection.The measurement results using the few-channel spectroscopy dose method showed that the PER could be controlled within±14%and±2%under the conditions of two and three energy intervals,respectively,with different channel numbers.The PER measured at 0°angle of radiation incidence meets the-29%to+67%requirements of IEC 61526:2010.Meanwhile,the channel number and counts-to-dose conversion factors formed in the experiment can be integrated into an EPD.
文摘在细粒度图像分类中,现有的小样本学习算法未能充分结合通道和空间信息提取细粒度图像的判别性特征,导致仅依靠单一类型的特征不足以准确捕捉细粒度对象的类间差异.针对这一难题,提出了一种基于通道先验感知的多尺度细化网络,旨在有效融合通道信息和空间信息,提升小样本细粒度图像分类的性能.通道先验感知模块实现了通道维度上注意力权重的动态分配,从而高效地捕捉通道先验信息;多尺度特征聚合过程充分利用细粒度图像中丰富的上下文信息,获取丰富的空间和边界细节特征;最后,特征细化模块对上述提取的通道和空间维度信息进行优化,实现了对关键区域的动态选择和强调,进而融合形成更精细、更具代表性的混合特征表示.所提算法在以Conv-4作为骨干网络时,在Stanford Cars、Stanford Dogs和CUB-200-2011三个细粒度数据集上的实验分类性能显著提升.在5 way 1 shot分类任务中,三个数据集的准确率分别达到了79.95%、66.97%和81.91%;在5 way 5 shot分类任务中,准确率则分别为93.42%、82.48%和93.19%.
文摘在小样本学习任务中,针对传统的骨干卷积网络在提取图像特征时,由于多层卷积忽视细节特征导致特征信息丢失,因而图像分类准确率不高的问题,提出了基于两阶段特征空间增强的小样本图像分类模型。首先,该模型在残差网络(residual network, ResNet)12的底层引入中值增强的空间和通道注意力块(median-enhanced spatial and channel attention block, MESC);然后,该模型在ResNet12的中高层引入空间组增强(spatial group-wise enhance, SGE)模块,提升卷积神经网络中的语义特征学习能力,使模型有效提取特征图关键信息。该模型通过增强有限的训练样本的特征表示来提高分类性能,增强模型对噪声的鲁棒性。结果表明,该模型在加州理工学院-加利福尼亚大学圣地亚哥分校鸟类(California Institute of Technology-University of California at San Diego birds, CUB)-200-2011数据集上,5类别1样本和5类别5样本2种参数设置下的分类准确率分别比分布传播图网络(distribution propagation graph network, DPGN)模型提高了约5.15%和1.92%;在分层图像网络(tiered ImageNet, tieredImageNet)数据集上,这2种参数设置下的分类准确率分别比DPGN模型提高了约1.04%和0.55%。该模型提升了小样本图像分类任务的性能。
基金Supported by the National Natural Science Foundation of China(61922075)the USTC Research Funds of the Double First-Class Initiative(YD2100002004).
文摘Background Electroencephalography(EEG)has gained popularity in various types of biomedical applications as a signal source that can be easily acquired and conveniently analyzed.However,owing to a complex scalp electrical environment,EEG is often polluted by diverse artifacts,with electromyography artifacts being the most difficult to remove.In particular,for ambulatory EEG devices with a restricted number of channels,dealing with muscle artifacts is a challenge.Methods In this study,we propose a simple but effective novel scheme that combines singular spectrum analysis(SSA)and canonical correlation analysis(CCA)algorithms for single-channel problems and then extend it to a few channel case by adding additional combining and dividing operations to channels.Results We evaluated our proposed framework on both semi-simulated and real-life data and compared it with some state-of-the art methods.The results demonstrate this novel framework's superior performance in both single-channel and few-channel cases.Conclusions This promising approach,based on its effectiveness and low time cost,is suitable for real-world biomedical signal processing applications.
