局部最小值

The key to feature extraction is to select the optimal feature combination from all the features. In this study, the optimal color feature combination,{/G,(R+4G+B)/6, H variance, S variance}, is selected using genetic neural network at first. Then the main texture features including the mean, variance, energy and entropy of the cooccurrence matrix, the secondary texture features including the maximum frequency number, absolute uniform, inertia and importance of the co-occurrence matrix are extracted.

在此基础上，提出适合多阈值和小目标图像分割的免疫遗传图像分割方法，该方法借鉴了生物免疫系统自调节、抗原识别和记忆功能，并结合了遗传算法对图像进行自适应区域的阈值分割，改善了遗传算法优化复杂问题时存在收敛速度慢和易陷于局部最优的不足，克服了传统阈值分割方法对目标大小的限制，是一种新型的有效的图像分割方法。

This paper applies GA-BP optimal algorithm to solve the problem of coordinate inversion in Robot's control.

但BP网络易形成局部极小值而得不到整体最优，缺乏全局搜索性。

The first one is integrated by the least squares and the radical basis function interpolation, which is aimed to the slight nonlinear sensors, and under the limited increament of computation, it can make the results more precise. The second is used for serious nonlinearity and based on the moving least squares which converts the approximation generally into locally, and the result of this way is also satisfied.

对于不是特别严重的非线性，采用最小二乘拟合与径向基函数残差插值进行融合重构，可以在增加有限计算量条件下提高数据的近似精度；对于非线性较严重的传感器，为兼顾局部特性，采用移动最小二乘法进行数据重构，它通过全局近似向局部近似的转化，同样使重构结果具有满意的近似精度。

LBIE, based on the local boundary equation, adopts the traditional moving least squares approximation which depends on only the values of the nodes in the domain of the problem or along its boundary. The whole process of integration is carried on over a local domain or its local boundary centered at the node in question. The local boundary equation can be rewritten to represent the values of the unknown function at the point of interest, and the essential boundary conditions can be directly and easily enforced by using the Green formula and the characters of the Dirac function.

它以局部边界积分方程为基础，采用移动最小二乘近似函数，从而只需要分布在问题域内及其边界上的节点的信息值，无需划分单元；整个积分是在以节点为中心的局部域及其边界上实现，所以不需要背景积分网格；借助于格林公式及Dirac函数的性质，将局部边界积分方程转化为所考虑点的未知函数的边界积分表达式，便于直接施加本质边界条件。

More than 30 kinds of meshfree methods are reviewed in this paper in the light of weighted residual method, and different meshfree methods can be viewed as different forms of weighted residual method and/or with different approximation functions. Various kinds of meshfree approximate schemes are presented in detail, including moving least square approximation, kernel and reproducing kernel approximation, partition of unity approximation, radial basis approximation, radial point interpolation and natural neighbor interpolation.

详尽介绍了各种无网格近似方案(包括移动最小二乘近似、核近似和重构核近似、单位分解近似、径向基函数近似、点插值近似、自然邻接点插值近似等)和无网格法中常用的各类加权余量法(伽辽金格式、配点格式、局部弱形式、加权最小二乘格式和边界积分格式等)，并讨论了数值积分方法和边界条件的处理等问题。

Aiming at the disadvantage of conventional BP neural network, such as selecting parameter values by the empirical method, slow convergence speed and easy trap into local minimum points, this paper designs an improved BP neural network system. In order to improve the network convergence rate and reduce the training error, this paper optimizes the initial connecting weight value of neural network by the use of ant colony algorithm and trains artificial neural network by Levenberg-Marquardt algorithm.

针对常规BP神经网络参数的经验式取值方法以及收敛速度慢，容易陷入局部最小点等缺陷，设计了一种改进的神经网络系统，利用蚁群算法优化神经网络连接权初值，并采用LM算法对人工神经网络进行训练，提高了网络的收敛速度，降低了训练误差。

The optimization method is an effective tool in back-analysis of displacements of roadway, choosing the bad iteration method will lead to many problems, such as slow constringent speed, bad stability of solutions, local minimization.

最优化方法是进行巷道位移反分析的有力工具，如果优化迭代方法选择不当，将导致迭代过程收敛缓慢、解的稳定性差、陷入局部极小值等问题，不能保证搜索到全局最优解。

The optimization method is an effective tool in back-analysis of displacements of roadway, choosing the bad iteration method will lead to many problems, such as slow constringent speed, bad stability of solutions, local minimization. The globally optimal solution cant be found out.

最优化方法是进行巷道位移反分析的有力工具，如果优化迭代方法选择不当，将导致迭代过程收敛缓慢、解的稳定性差、陷入局部极小值等问题，不能保证搜索到全局最优解。

By the modem time series analysis method, based on the ARMA innovation model, under the linear minimum variance optimal information fusion criterion, three distributed fusion steady-state optimal Kalman filters, predictors and smoothers weighted by matrices, scalars, and diagonal matrices are presented for multisensor systems with correlated input and observation noises, and with correlated observation noises. The Lyapunov equations and formulas of computing local filtering, predicting and smoothing error variances and covariances are given, which are applied to compute optimal weights. The corresponding three distributed fusion Wiener state estimators are also presented.

应用现代时间序列分析方法，基于自回归滑动平均新息模型，在线性最小方差最优信息融合准则下，对于带相关输入噪声和观测噪声和带相关的观测噪声的多传感器系统，提出了按矩阵加权、按标量加权和按对角阵加权的三种分布式融合稳态Kalman滤波器、预报器和平滑器，其中提出了局部滤波、预报和平滑估值误差方差阵和协方差阵的Lyapunov方程和计算公式，它们被用于计算最优加权，也提出了相应的三种分布式融合Wiener状念估值器。

The energy which drives the evolution of a curve is constructed based on local image information,the alignment term enhances the speed of curve evolution,and the improved data fitting term which is based on the competition between of the shifted Heaviside functions,makes the curve converge to a better global stationary minimum.

该算法基于图像局部化信息创建驱动曲线演化的能量，引入的配准项提高了曲线的演化速度，基于曲线演化竞争的数据拟合项，使得曲线能更稳定地收敛到一个全局静态最小值，且算法对水平集函数初始化位置不敏感。

Plunder melds and run with this jewel!

掠夺melds和运行与此宝石！

My dream is to be a crazy growing tree and extend at the edge between the city and the forest.

此刻，也许正是在通往天国的路上，我体验着这白色的晕旋。