算法

This conclusion can also be proved by spectrum-warping audio watermarking algorithm in MCLT domain. To discuss further deeply on why spectrum-warping technique does not have the required robustness, the studies are carried out upon the essentials of bilinear transformation. All the theoretical analysis and experiments results put the robustness of the spectrum-warping audio watermarking under questioning.

四、提出了基于量化的变换域音频水印算法中关于量化参数的选择方法量化变换系数的音频水印算法是以修改谱分量幅值为基础，算法的主要问题是如何选择量化参数以确保算法的鲁棒性和不可感知性，另外量化系数的区域选择也会影响算法的性能。

Now, many methods are used for the computing of latitude, ie, Heikkinnen arithmetic, Ozone arithmetic, polynomial approach arithmetic, ameliorative bowring arithmetic, and so on.

目前，在国内的纬度解算算法有很多，例如Heikkinen算法，Ozone算法，有理多项式逼近算法，以及改进的bowring算法等。

Second, according to model characteristic, such as non-linear, non-convex, multiple-peaked and so on, the paper discusses limitation of traditional solving algorithm, such as branch and bound method, GBD method and OA method, puts forward penalty function-makeup integer algorithm and genetic algorithm, analyses procedures of genetic algorithm, such as selection, crossover and mutation, brings forward modifications for specific realized problems of genetic algorithm, such as selection of initial points, operator design, adoptive value adjustment, constraint treatment and so on, and works out correlated solving program.

其次针对该模型非线性、非凸、多峰等特点，讨论了分支定界法、GBD法和OA法等传统求解算法的局限性，提出了罚函数一凑整算法和遗传算法的求解思路，对遗传算法的选择、交叉、变异等过程进行分析，对遗传算法初始点选择、算子设计、适应值调整和约束处理等具体实现问题提出了改进措施，并编制了相关的求解程序。

The dissertation studies thoroughly the problems of designing efficient AQMalgorithms. It proposes router buffer saturation compensator, delay estimation and compensation method, an adaptive framework and stability analysis.

论文对主动队列管理的算法设计进行了深入的研究，重点研究了AQM机制的性能分析、基于二阶最优模型的PI算法、缓冲区饱和补偿算法、AQM延时估测和补偿算法、AQM机制的参数自适应算法、AQM机制的稳定性和公平性。

The advanced research work of this paper is: we realize two kinds of DOA estimation arithmetic on the MATLAB software environment. One arithmetic is traditional arithmetic, which include Time-delay method and Capon method. The other one is MUSIC arithmetic. Then we qualitatively compare the performance of these arithmetic with each other.

论文的主要研究工作是：在MATLAB仿真软件平台上实现了DOA估计的传统算法(时延—相加算法和Capon最小方差算法)、多重信号分类算法MUSIC的仿真实现；并定性研究比较了三种算法的性能差别。

Finally, FastICA in time domain is derived based on instantaneous BSS system, and convolutional mixture BSS algorithm in frequency domain is obtained based on convolutional mixture BSS in time domain. Then we combine the two algorithms together and present the complex-value FastICA algorithm in frequency domain, which is a new BSS technique. We merge preprocessing in time domain and correlation coefficient solution in time domain into this new algorithm and present Time-Frequency domain FastICA BSS System.

最后，本文在瞬时混合盲分离系统基础上推出时域FastICA算法，在时域卷积混合盲分离系统基础上推出频域卷积混合盲分离算法，将这两种算法相结合，提出了频域复值FastICA算法，再将此算法与时域预处理方法以及最后回到时域中的利用相关系数求解分离信号方法相结合，从而建立了基于FastICA的时频域盲分离系统，并将其应用到实际环境中的语音信号盲分离中。

A costate prediction method based on bilinear model is presented. Using simple general model, two implemental DISOPElike algorithms based on Hopfield network and on temporal decomposition method are proposed respectively.

同时对于具有终端状态等式约束问题，建立了基于时变线性模型的递推DISOPE算法；对于非齐次双线性二次型问题提出了一个迭代算法，并且给出了该算法收敛的充分条件；给出了基于双线性模型的协状态预测算法；利用一般模型，分别提出了基于Hopfield网络和时间分解方法的DISOPE算法。

This approach not only inherits the advantages of absolute density based algorithms which can discover arbitrary shape clusters and are insensitive to noises , but also efficiently solves the following common problems: clustering results are very sensitive to the user-deflned parameters, reasonable parameters are hard to be determined, and high density clusters are contained fully in coterminous low density clusters. With this approach, incremental clustering can also be supported effectively by defining the affected sets and seed sets of the updating objects in this approach.

聚类算法的有效性问题主要表现在三个方面：其一，聚类算法大多要求用户输入一定的参数，例如希望产生的簇的数目，而这些参数通常难以确定，特别是针对高维空间中稀疏分布的实际应用数据集，用户几乎无法给出合适的算法参数，因此非专业用户需要与数据分析专家密切配合才能保证获得理想的聚类结果，导致算法的使用极为不便；其二，聚类结果对于输入的参数值过于敏感J，往往参数值的一些轻微变化却产生聚类结果的很大差异；其三，对于高维的实际应用数据集，其数据分布往往是稀疏的、杂乱的，很难为算法选择全局的参数进行准确的聚类分析，使得聚类的质量难以保证。

A new and efficient geometrical correction algorithm is proposed for MODIS IB data, and possesses of the merits of forward mapping and inverse mapping algorithms; 3. Algorithms to remove stripe noise in MODIS IB data are developed, including Fourier Transform, Wavelet Transform and interpolation method; 4. MODIS data composite algorithm is developed, and a 16-day composite map covering entire China and its outward is produced.

本文介绍了开发MODIS数据基础处理软件的基本算法和成果，主要包括：(1)用C++语言开发和实现了MODIS数据基础处理软件；(2)研究和发展了一个普适的、全新的、兼有前向映射和后向映射优点的MODIS 1B数据几何纠正算法；(3)研究和实现了MODIS条带噪声消除算法，包括傅立叶变换、小波变换和插值三种算法，并对其进行了比较；(4)研究和实现了MODIS影像合成算法，并在此基础上完成了中国及周围地区16天MODIS彩色影像和植被指数的合成。

Thirdly, in order to improve the learning efficiency and stability of feedforward neural network, a fast learning algorithm for neural networks based on CGM-OC approach is presented. Compared with other learning methods such as BP method, Broyden Flecher Goldfarl Shanno method and Powell method in which the derivative is replaced by the difference quotient etc, simulation results show that the proposed method is an efficient and fast one.

第三，为了提高前馈神经网络权值的学习效率及稳定性，提出一种基于正交校正共轭梯度优化方法的神经网络学习算法，通过与其它学习算法（如：BP算法、变尺度法、用差商近似代替导数的Powell法等）的比较，经仿真试验表明，本算法是一种高效、快速的学习算法。

For a big chunk of credit-card losses; the number of filings (and thus charge-off rates) would be rising again, whether

年美国个人破产法的一个改动使得破产登记急速下降，而后引起了信用卡大规模的亏损。

Eph. 4:23 And that you be renewed in the spirit of your mind

弗四23 而在你们心思的灵里得以更新