ID	原文	译文
53637	投放一定数量包含稀疏多源噪声信息的种子，通过限定噪声点在多源遥感中具有相同位置，实现多源对抗攻击的耦合，按种子信息制作对抗样本，	firstly, generate some random vectors that contains the information of multi-source adversarial perturbations and gen- erate the adversarial examples according to the vectors.
53638	利用上一代的种子( 父代) 进行变异与交叉，产生新的种子( 子代) ，同样制作对抗样本，	Then generates new parameter vectors by crossover operator and mutation operator and generate the adversarial examples, too.
53639	综合比较多源对抗样本的攻击效果，保留效果更好的种子，重复此过程，最终可得到高度耦合且攻击效果最好的多源遥感对抗样本。	Finally, evaluate the adversarial examples and choose the better vectors to survive. Repeat this progress and we can get the best multi-source adversarial examples.
53640	实验证明了本文方法的可行性:	Experimental results demonstrate the feasibility of our strategy.
53641	在单点攻击下，61. 38% 的光学图像和38. 93% 的合成孔径图像被成功转化为对抗样本，光学和合成孔径分类器中都无法正确识别的区域从 5. 83% 升至 55. 10% 。	Under one-pixel attack of our method, 61. 38% of optical images and 38. 93% of SAＲ images can be crafted to adversarial images.
53642	随着多模态数据的爆发式增长，跨模态检索作为一种搜索多模态数据的最常用方法，受到越来越多的关注。	With the explosive growth of multi-modal data, cross-modal retrieval, as the most commonly-used method to search multi-modal data, has received extensive attention.
53643	然而，目前存在的大多数深度学习的方法仅仅采用模型后端最后一个全连接层输出作为模态独有的高层语义表征，忽视了多个层次上不同尺度特征之间的语义相关性，具有一定的局限性。	However, most of the current deep learning methods only use the output of the final fully connected layer as the modal-special high-level semantic representation, ignoring the semantic correlation between features with different scales extracted from multiple levels, thus have certain limitations.
53644	为此，本文提出一种基于特征金字塔融合表征网络的跨模态哈希检索方法。	In this paper, we proposed a cross-modal hash retrieval method based on feature pyramid fusion representation network.
53645	该方法设计了一种特征金字塔融合表征网络，	This method designed a feature pyramid fusion representation network.
53646	通过在多个层次和不同尺度上进行特征提取并融合，挖掘多个层次上不同尺度下模态特征的语义相关性，充分利用模态特有的特征，使网络输出的语义表征更具有代表性。	Through feature extraction and fusion at multiple levels and different scales, the semantic correlation of modal features with different scales at multiple levels is mined, and the modal-special features are fully utilized to make the semantic representation of the network output more representative.