Kernel PCA与BP神经网络相结合的变压器故障诊断*

为了提高变压器故障诊断的准确率和抗干扰能力,提出一种基于核特征量的BP神经网络故障诊断模型。通过核主成分分析将故障样本从低维的特征空间非线性地映射到高维的核空间,提高了样本的可分性,然后以核特征量作为BP神经网络的输入特征量,建立变压器故障诊断模型。实验对比了结构相似、输入量不同的BP神经网络,结果表明采用核特征量的诊断模型具有更好的诊断效果和抗干扰能力。     

基于检测与跟踪相互迭代的极暗弱目标搜索算法

针对极低信噪比（LSNR）情况下暗弱运动目标和背景噪声的强度难以区分的问题,提出了一种基于检测与跟踪相互迭代的极暗弱目标搜索算法,总体上采用将时域检测与空域跟踪的过程联合、迭代进行的新型策略。首先,在检测过程中计算检测窗口内信号片段与已经提取的背景估计特征的差别;然后,在跟踪过程中运用动态规划算法保留使得轨迹能量累积最大的轨迹;最后,自适应地调整下一检测过程中被保留轨迹的检测器阈值参数,使该轨迹内的像素能以更宽容的策略被保留到下一检测跟踪阶段。实验测试结果表明,所提算法可以在1%~2%的虚警率和约70%的检测率下探测到低至0 dB的暗弱运动目标。可见该算法可有效改善对LSNR暗弱目标的检测能力。     

Research on adaptive notch filter cascading neural network anti-jamming algorithm

Satellite navigation suffers from performance degradation in the presence of narrowband interference. We present a Global Positioning System(GPS) receiver anti-jamming algorithm based on the neural network. The method combines the adaptive notch filter with the Back Propagation(BP) neural network. We first use the two order lattice infinite impulse response adaptive notch filter to eliminate-out-of-band jamming, then use the BP neural network predictor to estimate and cancel jamming. We analyze three aspects of algorithm performance which include acquired satellite numbers,signal-to-noise improvements and iteration numbers. Simulation results show that the method can effectively suppress narrowband interference and has a better anti-jamming capacity and a faster convergence speed.     

Global Avalanche Characteristics of Boolean Functions by Concatenation

In order to measure the correlation propeties of two Boolean functions,the global avalanche characteristics of Boolean functions constructed by concatenation are discussed,i.e.,f_1‖f_2and f_1‖f_2‖f_3‖f_4.Firstly,for the function f = f_1‖f_2,the cross-correlation function of f_1,f_2 in the special condition are studied.In this case,f,f_1,f_2 must be in desired form.By computing their sum-of-squares indicators,the crosscorrelation function between f_1,f_2 is obtained.Secondly,for the function g = f_1‖f_2‖f_3‖f_4,by analyzing the relation among their auto-correlation functions,their sum-of-squares indicators are investigated.Based on them,the sum-of-squares indicators of functions obtained by Canteaut et al.are investigated.The results show that the correlation property of g is good when the correlation properties of Boolean functions f_1,f_2,f_3,f_4 are good.     

一种新型Linux内核级Rootkit设计与实现

传统的内核级Rootkit无法绕过当前主流的Rootkit检测方法。对传统的内核级Rootkit的实现机制和当前主流的Rootkit检测方法进行了分析,设计并实现了一种新型Linux内核级Rootkit。通过实验证明,该Rootkit能够实现后门提权、进程隐藏及文件隐藏等功能,并能绕过当前主流的Rootkit检测工具的检测,其抗检测能力明显好于传统的内核级Rootkit。     

基于Intel VT硬件虚拟化的Rootkit技术

Intel VT硬件虚拟化技术使Rootkit可以利用底层优势实现深度隐藏。首先结合木马协同隐藏的思想,提出了基于Intel VT硬件虚拟化的Rootkit(HVRootkit)的协同隐藏模型,并给出形式化描述;然后根据该模型,在深入分析进程切换过程和操作系统内核数据结构的基础上,设计并实现了HVRootkit原型,该原型能够监控系统进程的切换过程,并通过修改与内核层进程视图和用户层进程视图相关的数据结构,隐藏系统进程。实验表明,HVRootkit原型符合协同隐藏的思想,能够实现对进程的深度隐藏,隐藏性能明显优于传统的内核级Rootkit。     

一种高效的全局K-均值算法

针对全局K-均值算法时间复杂度大的问题,提出一种增量选择初始聚类中心的新方法。选择数据集中周围分布最密集的样本作为第一个初始聚类中心,选择最小化目标函数贡献大,并且和已有聚类中心距离远的样本作为下一个初始聚类中心。改进算法减少了增量选取初始聚类中心时的计算量,降低了时间复杂度。实验证明,改进算法与全局K-均值算法、快速全局K-均值算法相比,在不影响聚类效果的基础上,减少了聚类时间,与优化初始聚类中心的算法相比,聚类效果更优。     

Hybrid Kalman and unscented Kalman filters for INS/GPS integrated system considering constant lever arm effect

In inertial navigation system(INS) and global positioning system(GPS) integrated system, GPS antennas are usually not located at the same location as the inertial measurement unit(IMU) of the INS, so the lever arm effect exists, which makes the observation equation highly nonlinear. The INS/GPS integration with constant lever arm effect is studied. The position relation of IMU and GPS’s antenna is represented in the earth centered earth fixed frame, while the velocity relation of these two systems is represented in local horizontal frame. Due to the small integration time interval of INS, i.e. 0.1 s in this work, the nonlinearity in the INS error equation is trivial, so the linear INS error model is constructed and addressed by Kalman filter’s prediction step. On the other hand, the high nonlinearity in the observation equation due to lever arm effect is addressed by unscented Kalman filter’s update step to attain higher accuracy and better applicability. Simulation is designed and the performance of the hybrid filter is validated.     

一种虹膜色素块检测与分类方法

针对现有虹膜识别系统中全局纹理特征提取方法忽略了纹理类型信息的问题,提出了一种针对全局性纹理中虹膜色素块的检测与分类方法.该方法利用灰度聚类法实现虹膜图像中色素块可能存在区域的初定位,依据坑洞和色素斑这两类色素块的灰度空间分布特性,定义一组区域特征参数作为分类特征向量,利用支持向量机实现二者的检测与分类.算法对图库中图像的坑洞和色素斑的检测正确率分别为99.2%和86.5%,对无特征纹理存在的虹膜图像检测正确率为87.2%.实验结果表明,该方法具有较高的正确率,能够满足虹膜识别系统的纹理特征提取要求.