ID 原文 译文
8774 进而分析AMP迭代次数对滤波对象结构特征与滤波算子性能的影响,研究AMP的阶段化滤波操作,提出一种基于卡通-纹理模型与分段滤波的AMP图像重构算法。 AMP is analyzed on the number of iterations to filter object structure characteristics and the influence of filter operator performance, the research of the AMP phase filtering operation, put forward a cartoon - based texture model with piecewise filtering AMP image reconstruction algorithm.
8775 实验表明,该算法能够更好地保留图像轮廓与纹理信息,提高图像的重构质量。 Experiments show that the proposed algorithm can better retain the image contour and texture information, improve the quality of image reconstruction.
8776 通过对合成孔径雷达(synthetic aperture radar,SAR)图像相干斑噪声的特点分析,提出一种基于贝叶斯模型的shearlet域SAR图像去噪方法。 Based on synthetic aperture radar (synthetic aperture radar, SAR) image, coherent noise characteristic analysis in this paper, a model based on bayesian shearlet domain of SAR image denoising methods.
8777 首先将变换后的SAR图像在shearlet域进行稀疏表示,得到稀疏系数的分布; Will first after the transformation of SAR image sparse representation in the shearlet domain to get the distribution of the sparse coefficient;
8778 其次利用贝叶斯模型进行信号和噪声检测的建模,得到最佳的阈值; Secondly by using bayesian model is used to detect the signal and noise model, get the best threshold value;
8779 然后根据稀疏系数在不同方向上相关性不同的特点,利用自适应加权收缩算法对SAR图像噪声进行平滑处理; Then according to the sparse coefficient of correlation of different characteristics in different directions, using the adaptive weighted shrinkage algorithm for SAR image smoothing noise;
8780 最后利用降噪后的高频子图像和低频子图像进行逆shearlet变换,得到SAR重构图像。 After the last use of noise of high frequency images and low frequency images inverse shearlet transform, SAR reconstruction images.
8781 通过在MSTAR数据库上的实验表明,该算法在滤除相干斑噪声的效果上比其他方法更好,并且不会损失图像的边缘特性。 By means of the experiment in MSTAR database show that the algorithm in the coherent noise in the filtering effect is better than other methods, on image edge features and not loss.
8782 为解决深度卷积神经网络(convolutional neural networks,CNN)难以训练的问题,提出一种快速、高效的双通道神经网络(dual-channel neural networks,DCNN)。 Convolution neural network to solve the depth (convolutional neural networks, CNN) hard training problems, puts forward a rapid and efficient dual channel neural network (dual channel - neural networks, DCNN).
8783 该神经网络由直通通道和卷积通道两种通道构成,直通通道负责保障深度网络的畅通性,卷积通道负责深度网络的学习。 The neural network consists of direct channel and convolution of two channels, direct channel is responsible for the security of deep web smooth general characteristic, convolution channel is responsible for the depth of network learning.