ID 原文 译文
16665 针对当前基于流形排序的显著性检测算法缺乏子空间信息的挖掘和节点间传播不准确的问题，该文提出一种基于低秩背景约束与多线索传播的图像显著性检测算法。 Considering the lack of subspace information digging and inaccurate propagation between nodes inexisting saliency detection algorithm based on manifold ranking, an image saliency detection algorithm based onbackground constraint of low rank and multi-cue propagation is proposed.
16666 融合颜色、位置和边界连通度等初级视觉先验形成背景高级先验，约束图像特征矩阵的分解，强化低秩矩阵与稀疏矩阵的差异，充分描述子空间结构信息，从而有效地将前景与背景分离； Primary visual priors such as color,location and boundary connectivity prior are fused to form a background high-level prior, which restrains thelow rank decomposition of feature matrix and strengths the difference between low rank matrix and sparesmatrix, describes structural information of subspace fully to separate foreground and background efficiently.
16667 引入稀疏感知和局部平滑等线索改进传播矩阵的构建，增强颜色特征出现概率低的节点的传播能力，加强局部区域内节点的关联性，准确凸显节点的属性，得到紧密且连续的显著区域。 Cues of rareness perception and local smoothing are introduced for improving the reconstruction of propagationmatrix, which improves the node’s propagation capacity that has low probability of color feature occurrence,enhances the relevance of local region, strengthens the properties of nodes accurately to obtain the compact andcontinuous salient regions.
16668 在3个基准数据集上的实验结果与图像检索领域的应用证明了该文算法的有效性和鲁棒性。 The experimental results on three benchmark datasets and the application to imageretrieval demonstrate the efficiency and robustness of the proposed algorithm.
16669 针对单一尺度卷积神经网络(CNN)对船舶图像分类的局限性，该文提出一种多尺度CNN自适应熵加权决策融合方法用于船舶图像分类。 Considering the limitation of single scale Convolutional Neural Network (CNN) for ship imageclassification, a self-adaptive entropy weighted decision fusion method for ship image classification based onmulti-scale CNN is proposed.
16670 首先使用多尺度CNN提取不同尺寸的船舶图像的多尺度特征，并训练得到不同子网络的最优模型； Firstly, the multi-scale CNN is used to extract the multi-scale features of shipimage with different sizes, and the optimum models of different sub-networks are trained.
16671 接着利用测试集船舶图像在最优模型上测试，得到多尺度CNN的Softmax函数输出的概率值，并计算得到信息熵，进而实现对不同输入船舶图像赋予自适应的融合权重； Then, the ship imagesof test set are tested on the optimum models, and the probability value that is output by Softmax function ofmulti-scale CNN is obtained, which is used to calculate the information entropy so as to realize the adaptiveweight assigned to different input ship images.
16672 最后对不同子网络的Softmax函数输出概率值进行自适应熵加权决策融合实现船舶图像的最终分类。 Finally, self-adaptive entropy weighted decision fusion is carriedout for the probability value that is output by Softmax function of different sub-networks to realize the finalship image classification.
16673 在VAIS数据集和自建数据集上分别进行了实验，提出的方法的分类准确率分别达到了95.07%和97.50%。 Experiments perform on VAIS (Visible And Infrared Spectrums) and self-builtdatasets respectively, and the proposed method achieves average accuracy of 95.07% and 97.50% on thesedatasets respectively.
16674 实验结果表明，与单一尺度CNN分类方法以及其他较新方法相比，所提方法具有更优的分类性能。 The experimental results show that the proposed method has better classificationperformance than those of the single scale CNN classification method and other state-of-the-art methods.