开放集识别(Open Set Recognition,OSR)的主要目的是识别未标记数据中的新类样本,同时对已见类样本进行正确分类.现有的大多数识别方法对未标记数据的评估和伪标记信息的利用不足.本文提出一种基于主动学习的开放集图像识别方法(Open Se...开放集识别(Open Set Recognition,OSR)的主要目的是识别未标记数据中的新类样本,同时对已见类样本进行正确分类.现有的大多数识别方法对未标记数据的评估和伪标记信息的利用不足.本文提出一种基于主动学习的开放集图像识别方法(Open Set Image Recognition Method Based on Active Learning,AC-OSIR),充分利用未标记数据提升开放集识别性能.通过引入已见类别的语义知识,构建语义知识和图像特征的映射关系.对于未标记数据,利用阈值选择策略区分开放集样本和已见类样本,通过主动学习模型迭代地识别高置信度开放集样本和已见类样本,并将高置信度已见类样本添加到标记数据集中.本文在图像分类数据集CIFAR-10、TIN和LSUN,以及两个合成数据集的实验结果表明了基于主动学习的开放集图像识别方法的有效性.展开更多
目的:探讨右心室主动电极起搏与被动电极起搏二者之间起搏参数的差异。方法:永久心脏起搏器患者186例,其中93例为右室间隔部主动电极起搏,93例为右室心尖部被动电极起搏,于植入时测试单极起搏阈值、感知及阻抗,所有患者的起搏脉宽均为0....目的:探讨右心室主动电极起搏与被动电极起搏二者之间起搏参数的差异。方法:永久心脏起搏器患者186例,其中93例为右室间隔部主动电极起搏,93例为右室心尖部被动电极起搏,于植入时测试单极起搏阈值、感知及阻抗,所有患者的起搏脉宽均为0.42ms。结果:有效起搏时主动电极组与被动电极组比较起搏阈值无显著差异(1.19±0.25 vs 0.98±0.25)V(P>0.05),被动电极组感知显著高于主动电极组感知(8.57±5.08 vs 7.69±4.21)mV(P<0.05),被动电极组阻抗显著高于主动电极组阻抗(718.60±239.50 vs 656.88±165.47)Ω(P<0.05)。结论:被动电极即刻感知优于主动电极,阻抗则高于主动电极。展开更多
A certain number of charges are deposited on the surface of high-voltage solar array because of effects of space plasma,high-energy charged particles,and solar illumination,hence the surface is charged.Phenomena of el...A certain number of charges are deposited on the surface of high-voltage solar array because of effects of space plasma,high-energy charged particles,and solar illumination,hence the surface is charged.Phenomena of electrostatic discharge(ESD) occur on the surface when the deposited charges exceed a threshold amount.In this paper,the mechanism of this ESD is discussed.The ground simulation experiment of the ESD using spacecraft material under surface charging is described,and a novel ESD protecting method for high-voltage solar array,i.e.an active protecting method based on the local strong electric field array is proposed.The results show that the reversal potential gradient field between the cover surface and the substrate materials of high-voltage solar array is a triggering factor for the ESD on the array.The threshold voltage for the ESD occurring on the surface is about 500 V.The charged particles could be deflected using the electric field active protecting method,and hence the ESD on the surface is avoided even when the voltage on the conductor array increases to a certain value.These results pave the way for further developing the protecting measures for high-voltage solar arrays.展开更多
In this paper, a novel mathematical morphological approach is proposed, which is combined with an active threshold-based method for the identification of morphological features from images with poor qualities. The alg...In this paper, a novel mathematical morphological approach is proposed, which is combined with an active threshold-based method for the identification of morphological features from images with poor qualities. The algorithm is very fast and needs low computing power.First, a mixed smooth filtering is designed to remove background noises. Second, an active threshold-based method is discussed to create a binary image to achieve rough segmentation. Third, some simple morphological operations, such as opening, closing, filling, and so on, are designed and applied to get the final result of segmentation. After morphological analysis, morphological features, such as contours, areas, numbers, locations, and so on, are obtained. Finally, the comparisons with other conventional methods validate the effectiveness, and an additional experimental result proves the repeatability of the proposed method.展开更多
Purpose Improvements in D'(the fatigability constant for running)subsequent to training interventions remain elusive.High-intensity interval training(HIIT)within the severe intensity domain for short durations(<...Purpose Improvements in D'(the fatigability constant for running)subsequent to training interventions remain elusive.High-intensity interval training(HIIT)within the severe intensity domain for short durations(<2-min)have been theorized to improve D'.The purpose of the present study was to assess this in a group of moderately trained individuals.Methods Eighteen participants completed graded exercise testing(GXT),40-m sprint testing and a 3-min all-out test(3MT)for running to determine key mechanistic and physiological parameters.Participants were randomly assigned into one of two groups based on intensity prescription(G140%=140%of critical speed[CS]),or time intervals(G90-s=90-s)to complete a twice-weekly training intervention for 6-weeks followed by re-assessment.Results No between-group differences were present either prior to or following the intervention.Substantial and meaning-ful improvements were detected during the post-intervention period for both groups for VO2max(G140%:+7.60%;G90-s:+11.67%),speed evoking VO2max(sVO2max;G140%:+4.33%;G90_s:+2.92%),gas exchange threshold(GET;G140%:+12.02%;G90-s:+20.52%),speed evoking GET(sGET;G140%:+4.17%;G90_s:+7.92%),CS(G140%:M=0.62 m/s;G90-s:M=0.46 m/s),D'(G140%:M=-56.34 m;G90-s:M=-18.36 m),FI%(G140%M=-6.75%;G90-s:M=-4.38%)and maximal distance(G140%:M=49.67 m;G90_s:M=58.38 m).Conclusions The prescribed intensities and durations were insufficient to elicit improvements in D'.Improvements in D'may be dependent on very short-duration intervals(i.e.<60 to 90-s)at speeds exceeding 140%CS but below maximal sprint speed.展开更多
文摘开放集识别(Open Set Recognition,OSR)的主要目的是识别未标记数据中的新类样本,同时对已见类样本进行正确分类.现有的大多数识别方法对未标记数据的评估和伪标记信息的利用不足.本文提出一种基于主动学习的开放集图像识别方法(Open Set Image Recognition Method Based on Active Learning,AC-OSIR),充分利用未标记数据提升开放集识别性能.通过引入已见类别的语义知识,构建语义知识和图像特征的映射关系.对于未标记数据,利用阈值选择策略区分开放集样本和已见类样本,通过主动学习模型迭代地识别高置信度开放集样本和已见类样本,并将高置信度已见类样本添加到标记数据集中.本文在图像分类数据集CIFAR-10、TIN和LSUN,以及两个合成数据集的实验结果表明了基于主动学习的开放集图像识别方法的有效性.
文摘目的:探讨右心室主动电极起搏与被动电极起搏二者之间起搏参数的差异。方法:永久心脏起搏器患者186例,其中93例为右室间隔部主动电极起搏,93例为右室心尖部被动电极起搏,于植入时测试单极起搏阈值、感知及阻抗,所有患者的起搏脉宽均为0.42ms。结果:有效起搏时主动电极组与被动电极组比较起搏阈值无显著差异(1.19±0.25 vs 0.98±0.25)V(P>0.05),被动电极组感知显著高于主动电极组感知(8.57±5.08 vs 7.69±4.21)mV(P<0.05),被动电极组阻抗显著高于主动电极组阻抗(718.60±239.50 vs 656.88±165.47)Ω(P<0.05)。结论:被动电极即刻感知优于主动电极,阻抗则高于主动电极。
基金Project supported by National Natural Science Foundation of China（51177173）, Elec- tromagnetic Environment Effect Key Laboratory Foundation（9140C87010313 JB34004）.
文摘A certain number of charges are deposited on the surface of high-voltage solar array because of effects of space plasma,high-energy charged particles,and solar illumination,hence the surface is charged.Phenomena of electrostatic discharge(ESD) occur on the surface when the deposited charges exceed a threshold amount.In this paper,the mechanism of this ESD is discussed.The ground simulation experiment of the ESD using spacecraft material under surface charging is described,and a novel ESD protecting method for high-voltage solar array,i.e.an active protecting method based on the local strong electric field array is proposed.The results show that the reversal potential gradient field between the cover surface and the substrate materials of high-voltage solar array is a triggering factor for the ESD on the array.The threshold voltage for the ESD occurring on the surface is about 500 V.The charged particles could be deflected using the electric field active protecting method,and hence the ESD on the surface is avoided even when the voltage on the conductor array increases to a certain value.These results pave the way for further developing the protecting measures for high-voltage solar arrays.
基金This research is partly supported by the innovative research fund of aerospace, research fund for the program of new century excellent talents in Heilongjiang provincial university No. 1155-ncet-008 and the Natural Science Foundation of Heilongjiang Province under grant No. QC2015084, F201132.
文摘In this paper, a novel mathematical morphological approach is proposed, which is combined with an active threshold-based method for the identification of morphological features from images with poor qualities. The algorithm is very fast and needs low computing power.First, a mixed smooth filtering is designed to remove background noises. Second, an active threshold-based method is discussed to create a binary image to achieve rough segmentation. Third, some simple morphological operations, such as opening, closing, filling, and so on, are designed and applied to get the final result of segmentation. After morphological analysis, morphological features, such as contours, areas, numbers, locations, and so on, are obtained. Finally, the comparisons with other conventional methods validate the effectiveness, and an additional experimental result proves the repeatability of the proposed method.
文摘Purpose Improvements in D'(the fatigability constant for running)subsequent to training interventions remain elusive.High-intensity interval training(HIIT)within the severe intensity domain for short durations(<2-min)have been theorized to improve D'.The purpose of the present study was to assess this in a group of moderately trained individuals.Methods Eighteen participants completed graded exercise testing(GXT),40-m sprint testing and a 3-min all-out test(3MT)for running to determine key mechanistic and physiological parameters.Participants were randomly assigned into one of two groups based on intensity prescription(G140%=140%of critical speed[CS]),or time intervals(G90-s=90-s)to complete a twice-weekly training intervention for 6-weeks followed by re-assessment.Results No between-group differences were present either prior to or following the intervention.Substantial and meaning-ful improvements were detected during the post-intervention period for both groups for VO2max(G140%:+7.60%;G90-s:+11.67%),speed evoking VO2max(sVO2max;G140%:+4.33%;G90_s:+2.92%),gas exchange threshold(GET;G140%:+12.02%;G90-s:+20.52%),speed evoking GET(sGET;G140%:+4.17%;G90_s:+7.92%),CS(G140%:M=0.62 m/s;G90-s:M=0.46 m/s),D'(G140%:M=-56.34 m;G90-s:M=-18.36 m),FI%(G140%M=-6.75%;G90-s:M=-4.38%)and maximal distance(G140%:M=49.67 m;G90_s:M=58.38 m).Conclusions The prescribed intensities and durations were insufficient to elicit improvements in D'.Improvements in D'may be dependent on very short-duration intervals(i.e.<60 to 90-s)at speeds exceeding 140%CS but below maximal sprint speed.
