速度函数

Modifications to the conventional layer by layer optimization method make it possible to avoid computing Hessian Matrix and its inverse form, thus it can decrease computational complexity dramatically. Simulation results for some practical problems such as function approximation and XOR problem, have demonstrated improvements over classical back-propagation algorithm and conventional layer by layer optimization algorithm.

本文在传统逐层优化算法上的改进避免了反复计算 Hessian 矩阵及其逆矩阵，因而可以减少计算量，对一些应用问题，如函数逼近、异或等问题的仿真实验证实了新的算法比经典的 BP 算法以及现存的逐层优化算法的训练速度更快、可靠性更好。

Key words: analytical mechanics; event space; variable mass; nonholonomic constraint; principle of Jourdain; Noether′s theorem

考虑N个质点组成的系统，其中每个质点的质量mi(i=1，…，N)是时间、位置、速度的函数， i为第i个质点mi的位矢。

For improving convergence, this paper firstly introduces Lagrange function to solve constraint problem in large-scale system optimization, secondly constructs the hierarchical optimization networks for large-scale industrial systems with Hopfield network.

为了 提高算法的收敛速度，引入Lagrange函数解决大系统优化问题中的各种约束，并用Hopfield 网络实现了大系统稳态递阶优化的网络算法，最后给出了某一大系统辨识及优化的仿真结果。

The satisfaction of Laplace equation of the velocity potential reduces the coefficients to only two unknown functions.

根据速度势满足拉普拉斯方程就可将这些系数简化到只有两个未知函数。

The principles are as follows:①preliminary interpretation is carried out on crosswell seismic data in depth domain to establish an accurate velocity model;②sinc interpolating function is used in depth-time conversion to realize waveform fidelity;③the recognition of surface seismic data is utilized to ascertain the reflection characteristics of crosswell seismic data in time domain;④ good spatial continuity of surface seismic data is adopted to retrieve the "hopper" defect in crosswell seismic profiles and the deficiency of lateral contrast between crosswell seismic data;⑤in fine calibration of well-side traces of crosswells and surface seismic data, coincident time-depth relation is used to match two seismic data so as to make the calibration more reasonable;⑥in multi-crosswell seismic data inversion by sparse spike inversion method, lateral restraints are strengthened to ensure the continuity of inversed wave impedance profile.

实现反演的基本原理是：对深度域井间地震资料进行初步解释；建立精确的速度模型；在对井间地震剖面进行深时转换时采用sinc插值函数实现波形保真；利用对地面地震资料的认识来落实时间域井间地震资料的反射特征；利用地面地震资料良好的空间连续性弥补井间地震剖面的"漏斗"缺陷和井间资料之间横向对比性的不足；在井间地震和地面地震井旁道精细标定中，通过一致的时深关系来实现两种地震资料之间的匹配，使井旁道的标定更加合理；在利用稀疏脉冲反演方法进行连井井间地震资料反演时，通过增强横向约束保证反演波阻抗剖面的连续性。

The forth algorithm is the same as the third one with an addition of finite termination criteria. Although the latter two algorithms have only linear rate of convergence, they are especially suitable for large-scale and sparse problems, with features of simple formula, small storage, sparsity preservation and easy implementation.

前者借助NCP函数把互补问题转化为等价的非光滑方程组，再用带参数的光滑方程组近似这些非光滑方程组，最后用牛顿型方法求解所得到的光滑方程组，希望通过光滑参数趋于零得到原来互补问题的解；后者基于等价不动点格式，构造了一个光滑迭代算法和一个具有有限终止性质的算法，虽然这种迭代算法仅有线性收敛速度，但由于其格式简单、存储量小、保稀疏性、非常易于计算机实现等特点，故较适用于求解大规模稀疏问题。

Both theoretical analysis and experimental results indicate that the linear dependence of the reciprocal of the time-correlation length on the velocity of the moving diffuser is still held in this case.

理论分析和实验测量结果均表明，这种动态散斑的空-时相关函数与一次散射形成的动态散斑相比，对散射体的运动十分敏感，但其时间相关长度与速度的倒数仍然保持线性关系。

The mentioned control scheme can effectively control two joint of space manipulator to stably track the desired trajectory in joint space. It has obvious advantages that with needless feedback and measured the position, velocity, acceleration, attitude angle velocity and attitude angle acceleration of the floating base. At the same time, no requirements for the dynamic equations of the system are linearly dependent on inertial parameters.

文中提到的控制方案能够有效地控制漂浮基空间机械臂的载体姿态及机械臂关节，可以协调地完成期望的轨迹运动，并具有不需要反馈和测量空间机械臂载体的位置、移动速度、移动加速度，同时也不要求系统动力学方程关于系统惯性参数呈线性函数关系的显著优点。

The paper first presents an objective model of task scheduling,and then based on the analysis of Q learning algorithm,the Markov decision process description of the scheduling problem is given.

首先建立任务调度问题的目标模型，在分析 Q学习算法的基础上，给出调度问题的马尔可夫决策过程描述；针对任务调度的 Q学习算法更新速度慢的问题，提出一种基于多步信息更新值函数的多步 Q学习调度算法。

In this paper, the theory, algorithm, and experiment of automatic object detection and tracking are studied in depth. It is firstly pointed out that the essential of Mean Shift method is a special Newton-Gaussian method. A new method named Fast Mean Shift is established to stretch the conservative step of Mean Shift method. The convergence and validity of this new method are proved in theory. And it is also proved that the convergence speed of Fast Mean Shift is faster than that of Mean Shift. The contrast experiments of searching the maximum possibility density of random of data sets in plane and 3D space are done. The results show that this new method can reduce the iterations greatly. A new object tracking method based on Fast Mean Shift is built to improve the object tracking performance, which is shown in the face tracking experiment with the tennis sequence provided by the Ohio State University, and the car tracking experiment with the car sequence provided by Kalsruhe University. The face trcking experiment with highly noised images show that the object tracking method based on Fast Mean Shift has strong anti-jamming ability. A new fast color object detection technology based on characteristic color is established, which use characteristic color distribution to compute the characteristic color vector of any area in an image quickly. With the high performance search method, the fast object detection is achieved. At last, using object tracker based on Fast Mean Shift and color object detector based on characteristic color with the Kalman filter, PID controller, searial communication and other technologies, automatic object detection and tracking system with control system is built. The availability and anti-jamming ability of this system are verified by some object detection and tracking tests in different scenes.

本文对目标自动识别与跟踪进行了理论、计算、试验三方面的深入研究，主要包括：首次指出了目标跟踪技术中常用的均值迁移方法的本质为一种特殊的高斯-牛顿方法，改进了均值迁移方法步长取值保守的弱点，建立了快速均值迁移方法，证明了该方法的收敛性、有效性以及收敛速度优于均值迁移方法；进行了平面和3维随机分布数据集的最大概率密度搜索对比试验，试验结果表明，快速均值迁移方法大大减少了迭代次数；建立了基于快速均值迁移的目标跟踪方法，利用俄亥俄州立大学提供的乒乓球序列图像和卡斯鲁厄大学的汽车序列图像，对人脸和汽车目标跟踪性能分别进行了对比试验，并进行了高噪声人脸图像目标跟踪试验，结果表明，基于快速均值迁移的目标跟踪方法有效提高了目标跟踪性能，具有很强的抗干扰能力；建立了一种新型彩色目标自动识别方法，采用特征色彩分布函数实现了对任意图像区域特征色彩矢量的快速计算，建立了高效的搜索方法，实现了彩色目标的快速识别；将基于快速均值迁移方法的目标跟踪方法、基于特征色彩的目标识别方法与卡尔曼滤波、PID控制、串行通讯等技术结合，建立了带有控制系统的快速目标自动识别与跟踪系统，并在不同场景下进行了目标自动识别与跟踪试验，验证了快速目标自动识别与跟踪系统的有效性和抗干扰能力。

In the negative and interrogative forms, of course, this is identical to the non-emphatic forms.

。但是，在否定句或疑问句里，这种带有"do"的方法表达的效果却没有什么强调的意思。

Go down on one's knees;kneel down

屈膝跪下。。。下跪祈祷