ID	原文	译文
4203	针对非负矩阵分解方法在有噪声的真实数据中获得特征的有效性问题，提出了一种稀疏诱导的流形正则化凸非负矩阵分解算法。	To address problems that the effectiveness of feature learned from real noisy data by classical nonnegative matrixfactorization method, a novel sparsity induced manifold regularized convex nonnegative matrix factorization algorithm(SGCNMF) was proposed.
4204	所提算法在流形正则化的基础上，向低维子空间的基矩阵添加基于 L 2,1 范数的稀疏约束，构建了乘法更新规则，分析在该规则下算法的收敛性，并设计了在低维子空间上不同噪声环境下的聚类实验。	Based on manifold regularization, the L 2,1 norm was introduced to the basis matrix of low dimen-sional subspace as sparse constraint. The multiplicative update rules were given and the convergence of the algorithm wasanalyzed. Clustering experiment was designed to verify the effectiveness of learned features within various of noisy envi-ronments.
4205	K均值聚类实验结果表明，稀疏约束降低了噪声特征在学习中的表达能力，所提算法在不同程度上优于同类 8 种算法，对噪声有更强的稳健性。	The empirical study based on K-means clustering shows that the sparse constraint reduces the representation ofnoisy features and the new method is better than the 8 similar algorithms with stronger robustness to a variable extent.
4206	针对声音信号受实际环境噪声影响引起耳蜗倒谱系数（CFCC）波动是导致击键内容识别率低的主要原因，研究相邻键 CFCC 之间的空间特征，建立基于点的 CFCC 空间梯度结构；	For the fluctuation of CFCC caused by environmental noise is the main reason for the low accuracy of key-stroke detection, the spatial characteristics of adjacent between CFCC were studied, and the spatial gradient structure ofCFCC based on points was established.
4207	在此基础上，在训练集和测试上研究 CFCC 空间梯度对击键内容识别的影响及确切邻域的点选取；	On this basis, the effect of CFCC spatial gradient on keystroke content recogni-tion and the selection of precise neighborhood points were studied on training and testing.
4208	最后，建立基于声音的高稳健性击键内容识别方法。	Finally, a high-robustness key-stroke recognition algorithm based on acoustic signals was constructed.
4209	在不同环境下进行的实验表明，所提 CFCC 空间梯度声音特征效果较好，识别准确率为 96.15% 。c	Extensive experiments in different environmentsdemonstrate that the proposed CFCC spatial gradient sound feature achieves great performance and the recognition accu-racy is 96.15%.
4210	针对脉内无意调相实现雷达辐射源个体识别时存在的分类模型性能不佳的问题，提出了一种长短时记忆加全卷积网络的雷达辐射源个体识别方法。	Aiming at the problem of poor performance of the classification model in the case of unintentional phase mod-ulation on pulse (UPMOP) to achieve radar specific emitter identification, a method for radar specific emitter identifica-tion with long and short-term memory and full convolutional networks (LSTM-FCN) was proposed.
4211	首先给出了脉内信号相位的简化观测模型，并对观测相位序列进行去斜处理，提取无意调相的含噪估计；	Firstly, a simplifiedobservation model of the intrapulse signal phase considering the intentional modulation was presented, and the observa-tion phase sequence was deramp to extract the noisy estimate of the UPMOP.
4212	然后利用贝塞尔曲线拟合无意调相，降低噪声的影响，获得无意调相更为精确的描述；	Then Bezier curve was utilized to fit theUPMOP to reduce the influence of noise and obtain a more accurate description of UPMOP.