ID 原文 译文
22785 在递归神经网络(RNN)语言模型输入中增加表示当前词所对应主题的特征向量是一种有效利用长时间跨度历史信息的方法。 Attaching topic features to the input of Recurrent Neural Network (RNN) models is an efficient method to leverage distant contextual information.
22786 由于在不同文档中各主题的概率分布通常差别很大，该文提出一种使用文档主题概率改进当前词主题特征的方法，并将改进后的特征应用于基于长短时记忆(LSTM)单元的递归神经网络语言模型中。 To cope with the problem that the topic distributions may vary greatly among different documents, this paper proposes an improved topic feature using the topic distributions of documents and applies it to a recurrent Long Short-Term Memory (LSTM) language model.
22787 实验表明，在 PTB 数据集上该文提出的方法使语言模型的困惑度相对于基线系统下降 11.8%。在 SWBD 数据集多候选重估实验中，该文提出的特征使 LSTM 模型相对于基线模型词错误率(WER)相对下降 6.0%；在 WSJ 数据集上的实验中，该特征使 LSTM 模型相对于基线模型词错误率(WER)相对下降 6.8%，并且在 eval92 测试集上，改进隐含狄利克雷分布(LDA)特征使 RNN 效果与 LSTM 相当。 Experiments show that the proposed feature achieved an 11.8% relatively perplexity reduction on the Penn TreeBank (PTB) dataset, and reached 6.0% and 6.8% relative Word Error Rate (WER) reduction on the SWitch BoarD (SWBD) and Wall Street Journal (WSJ) speech recognition task respectively. On WSJ speech recognition task, RNN with this feature can reach the effect of LSTM on eval92 testset.
22788 为有效提高雷达高分辨 1 维距离像目标识别系统的总体性能，需要对目标高分辨 1 维距离像进行特征提取，以得到具有最小信息损失、高可分性且低维度的目标特征，为实现该目的提出一种基于核主分量相关判别分析的特征提取算法。 For radar High Resolution Range Profile (HRRP) automatic target recognition, the features should be extracted with sufficient target information, high discrimination, noise robustness, and low feature vector dimension. However, radar HRRP recognition suffers from insufficient amount of information and low discrimination feature, besides, the radar recognition system also need the ability of real-time processing with low dimension. To obtain features with merits of low-dimension and high-discrimination, a novel feature extraction method is designed for radar high range resolution profile, namely Kernel Principal Component Correlation and Discrimination Analysis (KPCCDA).
22789 该算法基于目标高分辨 1 维距离像的统计特性，通过对核主分量分析中核函数的选择，实现对不同类型距离单元的特征提取。 With the proposed method, the statistical characteristics of different scatter range cells can be effectively used by Kernel Principal Component Analysis (KPCA).
22790 同时综合线性判别分析与典型相关分析理论构建新的准则函数，以实现特征空间中类内相关性与类间差异性最大化，同时减少目标特征中的冗余信息。 And the within-class correlation and between-class discrimination are maximized with linear discrimination analysis and canonical correlation analysis used. Besides, the redundancy and dimensionality of the feature vectors are reduced, yielding a lowered computational complexity to meet the storage requirement in practical radar target recognition.
22791 实验结果表明，该方法可以有效的提高目标高分辨 1 维距离像目标识别系统的总体性能。 Experimental results with measured data validate the efficiency of the proposed method.
22792 传统截面投影 Otsu 法后处理过程中的阈值 Q 为预先设定的常量，对含噪程度不同的图像普适性较差。 The threshold value of Q in the post process of traditional cross section projection Otsu's method is a constant, which is not universal applicability for images with different noises.
22793 该文提出一种基于记忆分子动理论优化算法的多目标截面投影 Otsu 法。 To solve this problem, this paper proposes a multi-objective cross section projection Otsu's method based on memory knetic-molecular theory ptimization algorithm.
22794 该方法将阈值 Q 作为变量，结合分割阈值 T，基于最大类间方差和最大峰值信噪比准则建立多目标图像分割模型，以兼顾图像分割的准确性和抗噪性； Based on the maximum between-class variance criterion and the maximum Peak Signal to Noise Ratio (PSNR) criterion, a multi-objective image segmentation model is established to take into account the segmentation accuracy and anti-noise capability for image segmentation by combining threshold Q with segmentation threshold T.