ID	原文	译文
5744	为提高极化合成孔径雷达（synthetic aperture radar,SAR）图像变化检测算法的鲁棒性以及检测精度,提出基于范数主成分分析（linorm principal component analysis)	To improve the polarization synthetic aperture radar (synthetic aperture radar, SAR) image change detection algorithm robustness and accuracy, is put forward based on the norm principal component analysis (linorm principal component analysis)
5745	为解决双势阱Duffing-van der Pol振子同频微弱信号定量检测问题,分析发现了Poincare截面能明显区分系统不同状态的现象,构建了基于Poincare截面的检测统计量,并以此为基础设计了信号检测的有效实现方法。	In order to solve the double trap Duffing - van der Pol oscillator frequency weak signal detection problem, quantitative analysis showed that the Poincare section can clearly distinguish between different system states phenomenon, based on the Poincare section of the test statistics is constructed, and on this basis, design the effective implementation of signal detection method.
5746	首先对Poincare截面进行数值计算得到系统关于策动力幅值的分岔点图,其次对不同状态下各幅值对应分布点做方差统计构建检测统计量。	Numerical calculation was carried out on the Poincare section first get system about ce power amplitudes of the bifurcation point figure, next to the corresponding amplitude distribution points under different state variance statistics to construct test statistics.
5747	不同状态的检测统计量分布区间不同,按照统计值落入的区间即可判断信号有无。	The distribution range is different, different states of statistics according to the statistics in the interval can be any judgment signal.
5748	仿真实验给出了不同策动力角频率下检测统计量的临界值和检测区间,并通过与相平面检测结果的对比验证了该检测方法的可行性,为微弱信号定量检测提供了参考。	Simulation experiment shows the different threshold policy under dynamic angular frequency statistics and testing interval, and compared with testing results of phase plane verify the feasibility of the method, the reference for the quantitative detection for weak signal.
5749	更快速区域卷积神经网络(faster region-based convolutional neural network,Faster RCNN)是两阶段的目标检测模型,通过区域生成网络将区域提议与识别完全融合到网络模型中,使主要的运算可以在图形处理器中完成,因此,其同时具有良好的检测速度与精度。	Faster regional convolution neural network (faster - region - -based convolutional neural network, the faster - RCNN) is the target detection model of two phase, through regional generation network will be proposed to the network and identify completely fusion model, make the main operation can be done in the graphics processors, therefore, its also has good detection speed and precision.
5750	但是当Faster RCNN直接应用于遥感图像目标检测,面对宽尺寸范围的多种目标时,性能受到了很大削弱。	But when Faster - RCNN directly applied to remote sensing image target detection, face the wide size range of a variety of goals, performance is much weakened.
5751	分析了池化操作和目标尺寸对区域提议的影响,提出联合多层次特征进行区域提议的方法,提升了目标区域的提议召回率。	Pooling operation is analyzed and the target size influence on regional proposal, the proposed method of joint multi-level features, improve the recall rate target area for proposal.
5752	针对性地优化前景样本的生成策略,避免训练过程中的产生无效前景样本,使得整个检测模型的训练更加高效。	The generate optimized prospect samples according to the strategy, avoid produce invalid prospects in the process of training samples, make the whole testing model of training more efficient.
5753	实验结果表明,所提出的模型和训练方法能够提高多尺度遥感图像目标的召回率与检测精度,且具备较高的训练效率。	The experimental results show that the proposed model and training method can improve the multi-scale recall rate and precision of remote sensing image target, and have higher training efficiency.