嵌入式隐Markov模型的分段训练方法

针对嵌入式隐Markov模型再学习问题,提出了分段训练方法用于人脸识别:把当前的训练样本看作整体训练样本的一部分,训练结束后存储训练后的模型参数和中间变量;增加新样本后,以当前模型参数作为初始模型参数,用新增样本训练模型,得到新的中间变量,最后将已存储的中间变量和用新样本计算得到的中间变量合成,得到最终的模型.人脸识别实验结果表明了该方法的可行性.      

基于模糊支持向量机的飞机飞行动作识别

传统的支持向量机由两类扩展到多类问题时,出现不可分区域。针对飞行动作识别提出解决这一现象的模糊支持向量机。采用模糊支持向量机对某型飞机飞行动作进行识别。实际飞参数据(6种飞行动作模式)的识别结果表明,模糊支持向量机较传统的多类支持向量分类器在飞机飞行动作识别率上有显著提高。     

航空发动机神经网络内模控制

基于神经网络内模控制理论研究了神经网络内模控制方法在航空发动机控制系统中的应用,建立了基于Elman网络的航空发动机多变量内模控制系统。采用Elman网络建立发动机内模型和内模控制器,详细介绍了建模、控制的算法及其实现过程。取部分飞行条件下的数据作为学习样本,采用动态BP算法对神经网络权值进行调整,并在飞行包线内其它工作点对整个控制系统进行了仿真。结果表明,使用神经网络建立的航空发动机内模控制系统具有良好的控制品质和较强的自适应能力。      

Obstacle avoidance handling and mixed integer predictive control for space robots

This paper presents a novel obstacle avoidance constraint and a mixed integer predictive control (MIPC) method for space robots avoiding obstacles and satisfying physical limits during performing tasks. Firstly, a novel kind of obstacle avoidance constraint of space robots, which needs the assumption that the manipulator links and the obstacles can be represented by convex bodies, is proposed by limiting the relative velocity between two closest points which are on the manipulator and the obstacle, respectively. Furthermore, the logical variables are introduced into the obstacle avoidance constraint, which have realized the constraint form is automatically changed to satisfy different obstacle avoidance requirements in different distance intervals between the space robot and the obstacle. Afterwards, the obstacle avoidance constraint and other system physical limits, such as joint angle ranges, the amplitude boundaries of joint velocities and joint torques, are described as inequality constraints of a quadratic programming (QP) problem by using the model predictive control (MPC) method. To guarantee the feasibility of the obtained multi-constraint QP problem, the constraints are treated as soft constraints and assigned levels of priority based on the propositional logic theory, which can realize that the constraints with lower priorities are always firstly violated to recover the feasibility of the QP problem. Since the logical variables have been introduced, the optimization problem including obstacle avoidance and system physical limits as prioritized inequality constraints is termed as MIPC method of space robots, and its computational complexity as well as possible strategies for reducing calculation amount are analyzed. Simulations of the space robot unfolding its manipulator and tracking the end-effector’s desired trajectories with the existence of obstacles and physical limits are presented to demonstrate the effectiveness of the proposed obstacle avoidance strategy and MIPC control method of space robots.     

基于模型重构的变形叶片配准定位技术

叶片毛坯的精确定位是叶片数控加工的核心问题,为了提高叶片数控加工质量,保证叶片各处余量分配均匀,本文以叶片进排气边的自适应数控加工中的检测定位为研究对象,提出了一种基于模型重构的配准定位方法.该方法利用偏置模型的叶片测量方式提高了测量精度,对基于重构的测量模型进行变形与定位分析,得出较优的定位参数.通过仿真试验和对比分析,论证了本文方法的正确性.     

基于模型大扭角叶片的三维矢量测量技术

针对接触式坐标测量机测量大扭角叶片截面存在余弦误差的问题，提出了基于模型的矢量测量方法。本文对此方法的测量精度及测量效率等方面进行论述，并通过实例验证了基于模型的矢量测量方法在精度与效率等方面优势，实现了此测量方法的工程化应用。     

A robust star identification algorithm with star shortlisting

A star tracker provides the most accurate attitude solution in terms of arc seconds compared to the other existing attitude sensors. When no prior attitude information is available, it operates in “Lost-In-Space (LIS)” mode. Star pattern recognition, also known as star identification algorithm, forms the most crucial part of a star tracker in the LIS mode. Recognition reliability and speed are the two most important parameters of a star pattern recognition technique. In this paper, a novel star identification algorithm with star ID shortlisting is proposed. Firstly, the star IDs are shortlisted based on worst-case patch mismatch, and later stars are identified in the image by an initial match confirmed with a running sequential angular match technique. The proposed idea is tested on 16,200 simulated star images having magnitude uncertainty, noise stars, positional deviation, and varying size of the field of view. The proposed idea is also benchmarked with the state-of-the-art star pattern recognition techniques. Finally, the real-time performance of the proposed technique is tested on the 3104 real star images captured by a star tracker SST-20S currently mounted on a satellite. The proposed technique can achieve an identification accuracy of 98% and takes only 8.2?ms for identification on real images. Simulation and real-time results depict that the proposed technique is highly robust and achieves a high speed of identification suitable for actual space applications.     

黎曼几何在计算机视觉中的应用研究进展

在计算机视觉领域，许多任务相关数据具有非欧结构，近年来基于黎曼几何的数据表征与应用研究受到了越来越多的关注。如何充分利用数据的几何结构，来提高目标识别、目标跟踪及目标检测算法的性能，是其中的一些研究热点。本文主要从三个方面介绍黎曼流形在计算机视觉中的应用研究进展。首先，从数学基本概念出发，阐述黎曼流形与图像的关系以及视觉应用的可行性，并介绍计算机视觉中具有重要应用的几种黎曼流形。其次，对黎曼流形在计算机视觉中若干常见应用进行了概述，重点介绍了与深度学习相结合的相关进展。最后，对引入黎曼流形的机器学习方法的未来发展进行了分析和讨论。     

Weighted marginal fisher analysis with spatially smooth for aircraft recognition

Due to limitations to extract invariant features for recognition when the aircraft presents various poses and lacks enough samples for training, a novel algorithm called Weighted Marginal Fisher Analysis with Spatially Smooth （WMFA-SS） for extracting invariant features in aircraft rec- ognition is proposed. According to the Graph Embedding （GE） framework, Heat Kernel function is firstly introduced to characterize the interclass separability when choosing the weights of penalty graph. Furthermore, Laplacian penalty is applied to constraining the coefficients to be spatially smooth in this algorithm. Laplacian penalty is able to incorporate the prior information that neigh- boring pixels are correlated. Besides, using a Laplacian penalty can also avoid the singularity of Laplacian matrix of intrinsic graph. Once compact representations of the images are obtained, it can be considered as invariant features and then be performed in classification to recognize different patterns of aircraft. Real aircraft recognition experiments show the superiority of our proposed WMFA-SS in comparison to other GE algorithms and the current aircraft recognition algorithm; the accuracy rate of our proposed method is 90.00% for dataset BH-AIR1.0 and 99.25% for dataset BH-AIR2.0.     

Multi-view space object recognition and pose estimation based on kernel regression

The application of high-performance imaging sensors in space-based space surveillance systems makes it possible to recognize space objects and estimate their poses using vision-based methods. In this paper, we proposed a kernel regression-based method for joint multi-view space object recognition and pose estimation. We built a new simulated satellite image dataset named BUAA-SID 1.5 to test our method using different image representations. We evaluated our method for recognition-only tasks, pose estimation-only tasks, and joint recognition and pose estimation tasks. Experimental results show that our method outperforms the state-of-the-arts in space object recognition, and can recognize space objects and estimate their poses effectively and robustly against noise and lighting conditions.