ID 原文 译文
7494 以层次结构构造一个集特征提取和分类器训练于一体的端到端网络,同时利用深度网络的反向传播完成分类器对特征提取的反馈以改进特征的效果。 With hierarchy construction a set of feature extraction and classifier training in an end-to-end network, at the same time using the depth of back propagation network to complete the classifier for feature extraction of feedback to improve the effect of the characteristics.
7495 在仿真模拟实验上,该网络框架分类目标达到了78.61%的精度,与其他方法相比,有效提高了目标的识别精度。 In simulation experiment, the network framework classification target accuracy reached 78.61%, compared with other methods, effectively improve the identification precision of the goal.
7496 所提框架能有效分类识别水下目标,具有良好的识别精度,且具备模块化结构,无需复杂预处理,实现简单。 The proposed framework can effectively underwater target classification recognition, good recognition accuracy, and possess modular structure, without complex pretreatment, implementation is simple.
7497 显著性检测是指自动提取未知场景中符合人类视觉习惯的兴趣目标的方法。 Significant testing refers to the automatic extraction of unknown scenario to conform to the goal of human visual habits interest method.
7498 为了进一步提高检测的有效性,同时降低像素类检测算法的计算量和复杂度,提出了基于动态指导滤波的图像显著性检测方法。 In order to further improve the effectiveness of the detection, at the same time reduce the pixel class testing calculation and complexity of the algorithm, is proposed based on dynamic detection of significant guiding filter images.
7499 在新设计的简单迭代指导滤波中,核函数不再像经典指导滤波器那样只利用固定的指导图像,而是利用了输入图像和动态指导图像的联合结构信息,它保证了指导图像对原输入图像较好的结构传递性。 In the new design of simple iterative guidance filter, kernel function is no longer just like classic guide filter using fixed image, the guidance of it using the joint structure of the input image and dynamic guidance image information, it ensures the guiding image good transitivity structure of the original input image.
7500 其次,为了节约算法的时间成本,采用采样的方式降低算法计算中需要的计算量。 Secondly, in order to save time cost of the algorithm and with the method of sampling to reduce computation algorithm in computing needs.
7501 最后,为了提取更有效地的显著性区域,引入了关键显著性区域提取方法,通过修正关键点集合得到更准确的目标区域。 Finally, in order to extract more effectively significant area, introduces the key significant region extraction method, through the fixed point set to get the more accurate the target area.
7502 实验结果表明,相比于其他像素类的显著性检测方法,该算法可以更快速和有效地检测出显著性目标。 The experimental results show that, compared to other pixel class the significance of detection method, this algorithm can more quickly and effectively detect the significant goal.
7503 基于信息流提出贝叶斯网络结构学习的改进型搜索评分算法。 Based on the information flow of bayesian network structure learning advanced search scoring algorithm.