ID 原文 译文
22565 全局建模缺少单幅图像结构信息，同时此类参数方法受到复杂度限制，无法使用大规模的弱监督训练数据。 This method lack of structure information in single image and suffer from enormous quantity parameters which result in expensive computation.
22566 针对以上问题，该文提出一种基于纹元森林和显著性先验的弱监督图像语义分割方法。 In this study, a weakly-supervised semantic segmentation algorithm is proposed.
22567 算法使用弱监督数据和图像显著性训练随机森林分类器用于语义纹元森林特征(Semantic Texton Forest, STF)的提取。 Specifically, a nonparametric random Semantic Texton Forest (STF) is obtained using weakly supervised training data and images saliency.
22568 测试时，先将图像进行过分割，然后提取超像素语义纹元特征，利用朴素贝叶斯法进行超像素标记的概率估计，最后在条件随机场(CRF)框架下结合图像显著性信息定义了新的能量函数表达式，将图像的标注(labeling)问题转换为能量最小化问题求解。 Under Conditional Random Field (CRF) framework, an novel energy function expression is developed based on saliency priors as structure context relationship, which avoids the construction of a huge graph in whole training dataset. Then STF feature is extracted from image superpixels and probability estimates of superpixels label is calculated by naive Bayesian method. Finally, a CRF-based optimization algorithm is proposed which can be efficiently solved by alpha expansion algorithm.
22569 结果表明，在并未对整个训练集建立图模型的情况下，仅利用单幅图像的显著性信息也可以得到较好的分割结果，同时非参模型有利于规模数据分析。 Experiments on the MSRC-21 dataset show that the new algorithm outperforms some previous influential weakly-supervised segmentation algorithms with no building graph in whole training set.
22570 基于常规时频分析方法的跳频信号参数估计中，采用核函数抑制时频分布交叉项会导致时频聚集性的下降，不利于信号参数提取。 In the case of parameter estimation of Frequency Hopping (FH) signal based on conventional time-frequency analysis, the suppression of cross-terms in time-frequency Distribution (TFD) by kernel function always leads to the decrease of time-frequency concentration, which is adverse to signal parameter extraction.
22571 针对此问题，该文提出一种基于稀疏时频分布(STFD)的跳频信号处理方法。 To deal with this problem, a kind of Sparse TFD (STFD) based FH signals processing method is proposed.
22572 该方法首先根据 Cohen 类分布的原理和跳频信号模糊函数的特点，以模糊域矩形窗为核函数，构建了一种 Cohen 类的矩形核分布(RKD)。 Based on the principle of Cohen's class of TFD and the ambiguity function characteristics of FH signals, a Rectangle-shaped Kernel Distribution (RKD) is constructed by choosing the rectangle function in ambiguity domain as its kernel function.
22573 RKD 可有效抑制交叉项，但其时频分辨率较低。 RKD can suppress the cross-terms effectively but is followed by poor time-frequency resolution.
22574 为提高 RKD 的时频性能，在压缩感知框架下，利用跳频信号时频分布的稀疏特性，对 RKD 附加稀疏性约束，建立稀疏时频分布(STFD)的优化求解模型。 In order to improve the performance of RKD, the TFD sparsity of FH signals is analyzed and utilized, and the optimal model of STFD is established by additional constraints to RKD under the Compressed Sensing (CS) frame.