特征协同与细粒度感知的遥感图像小目标检测

针对遥感图像中小目标多、排列密集导致的漏检问题,提出一种特征协同与细粒度感知的遥感图像小目标检测算法。首先,构造精细特征协同策略,通过智能调整卷积核参数,优化了特征间的交互和整合过程;通过精确控制信息流,实现从粗糙到精细的渐进式特征精化。在此基础上,本文设计一个细粒度感知模块,将感知注意力与移动反向卷积结合形成一个增强型检测头,显著增强网络对于极小尺寸物体的感知能力。最后,为了提升模型训练的效率,采用MPDIoU和NWD作为回归损失函数,解决位置偏差,加快模型收敛。在DOTA1.0数据集和DOTA1.5数据集上的实验结果表明,改进后算法相比于基准方法,平均精度分别提高7.4%和6.1%,相较于其他算法具有明显优势,显著改善遥感图像中小目标的漏检情况。

基于LSTM的多维度进程异常行为检测

恶意进程是威胁系统安全的一个重要因素,通过检测系统调用序列能有效发现恶意进程。不同的进程类型产生的调用序列会出现长度等方面的差异,因此单一化的检测方法无法在多种进程类型的检测上保持较高的准确率。针对这个问题,提出了基于长短时记忆(Long Short-Term Memory,LSTM)神经网络的多维度进程异常行为检测方法,在时间维度的基础上通过N-gram算法增加数据的信息维度,在多维数据形式中选取特征表现更好的维度进行异常判决。在UNM和ADFA-LD数据集上的实验结果表明,多维度的方法可以丰富调用序列的特征表现,以此减小不同进程类型的特征差异,且在多种进程类型下均有较好检测效果,使检测泛化能力得到提升。该方法在ADFA-LD数据集上与常用机器学习模型相比,准确率比支持向量机(Support Vector Machine,SVM)高11个百分点且误报率降低了70%,且相对于其他模型,检测效果均有所提升。

高性能PtS_(2)/MoTe_(2)异质结红外光电探测器

由于光电探测器的工作性能直接关系到系统数据采集质量,为此,对高性能PtS_(2)/MoTe_(2)异质结红外光电探测器进行了研究。通过选取材料、试剂和设备制作了PtS_(2)/MoTe_(2)异质结红外光电探测器。搭建探测器性能测试环境,并利用光响应度、探测率、响应时间和光电导增益4个指标,分析探测器性能。结果表明,随着测试时间的推移,PtS_(2)/MoTe_(2)异质结红外光电探测器的光响应度数值始终处于5 A/W限值以上;无论对采集何种材质反射的红外光,探测器探测率均大于10 cm·Hz1/2 W^(-1);无论光生电流是处于上升还是下降时间,其响应时间始终在限值150μs以下;光电导增益值保持在80%以上。

无线电波发射功率防爆要求与检测方法

现行国家标准GB/T 3836.1—2021《爆炸性环境第1部分:设备通用要求》和国际标准IEC 60079-0:2017《Explosive atmospheres-Part 0:Equipment-General requirements》规定无线电发射器的阈功率为无线电发射器的有效输出功率与天线增益的乘积。在无线电波防爆安全发射功率阈值一定的条件下,天线增益越大,无线电发射器的有效输出功率就越小,这将限制通过增大天线增益,提高无线传输距离。因此,有必要对国家标准GB/T 3836.1—2021和国际标准IEC 60079-0:2017中规定的阈功率的正确性进行研究,提出合理的无线电波发射功率防爆要求与检测方法。提出了无线电波防爆安全发射功率与天线增益无关,国家标准GB/T 3836.1—2021和国际标准IEC 60079-0:2017规定的无线电发射器的阈功率是错误的。提出了煤矿井下无线电波防爆安全发射功率阈值应大于16 W,且与天线增益无关;国家标准GB/T 3836.1—2021和国际标准IEC 60079-0:2017规定阈功率不得大于6 W,且与天线增益相关,是错误的。提出了无线电波防爆安全性能检测方法——检测无线电发射器输出功率,这既可保证通过检测的防爆无线电设备的防爆安全,又简化了检测方法,更提高了防爆无线电设备的无线电波发射功率,解除了对天线增益的限制,将大大提高煤矿井下防爆无线电设备的无线传输距离。

融合稀疏注意力机制在DDoS攻击检测中的应用

针对现有的DDoS(distributed denial of service)攻击检测模型面临大量数据时,呈现出检测效率低的问题。为适应当前网络环境,通过研究DDoS攻击检测模型、提取流量特征、计算攻击密度,提出一种基于融合稀疏注意力机制的DDoS攻击检测模型GVBNet(global variable block net),使用攻击密度自适应计算稀疏注意力。利用信息熵以及信息增益分析提取攻击流量的连续字节作为特征向量,通过构建基于GVBNet的网络模型在两种数据集上进行训练。实验结果表明,该方法具有良好的识别效果、检测速度以及抗干扰能力,在不同的环境下具有应用价值。

一种适用于空间相关信道的分集接收方法

针对空间相关信道下信号检测误码率性能下降的问题,提出一种多天线分集接收方法。该方法通过增加接收机天线个数,减小信道空间相关性对信号检测的影响,可根据实际工程需求灵活配置相应的接收天线数量,在避免引入高复杂度检测算法的同时具有较优的检测误码率性能。采用等增益合并方式对所提方法进行仿真分析,仿真结果验证了方法的有效性。

P-型核磁共振成像测井仪GAIN值测量故障分析

MRIL-Prime是哈里伯顿公司新一代的核磁共振成像测井仪器,通常简称为P-型核磁。在应用中发现无增益、增益不稳等常见故障。在实际维修中,必须掌握P-型核磁共振成像测井仪GAIN值的测量原理及中心频率的确定方法。当核磁探头通断绝缘都正常的情况下,增益值不稳或为零,应先检查线路中天线接口模块A2板上的开关管有无损坏,通常情况下增益值不稳或没有,多为A2板上的开关管损坏,更换即可。

面向钓鱼网站敏感特征项选取的IIGAIN算法

传统的钓鱼网站检测技术主要采用随机或者凭经验选取敏感特征项用于检测的方法,无法保证检测的准确性。为此,提出一种面向钓鱼网站敏感特征选取的改进的信息增益算法IIGAIN(Improved Information Gain Algorithm)。该算法综合考虑了特征项的类内离散度,通过对特征项的类内离散度差值做相应的处理,以处理后的结果作为惩罚项改进信息增益算法。实验结果表明,利用IIGAIN进行特征项选取的钓鱼网站检测方法的检测准确性明显优于随机选取特征项的钓鱼网站检测方法。

Ultrasonic flaw detection of discontinuous defects in magnesium alloy materials

Phased array ultrasonic testing, an effective ultrasonic testing(UT) technology, has been widely used in steel inspection because of its high accuracy, sensitivity, and efficiency. However, as its application in as-cast magnesium alloys has just begun, more research is needed. Considering the important role of the gain compensation in quantifying defects in magnesium alloys by ultrasonic phased array technology, the effects of microstructure, the position, size, and overlap of defects, and boundary distance(distance from the defect position to the side surface of the test casting) on gain compensation of as-cast AZ80 and AZ31 magnesium alloys were studied. Results show the gain compensation increases with the increase of grain size. There is a strict linear positive correlation between gain compensation and defect depth, but such relationship no longer exists due to the defects overlap, orientation and boundary distance. In addition, there is a strict linear negative correlation between the gain compensation and defect size.

Cued search algorithm with uncertain detection performance for phased array radars

A cued search algorithm with uncertain detection performance is proposed for phased array radars. Firstly, a target search model based on the information gain criterion is presented with known detection performance, and the statistical characteristic of the detection probability is calculated by using the fluctuant model of the target radar cross section （RCS）. Secondly, when the detection probability is completely unknown, its probability density function is modeled with a beta distribution, and its posterior probability distribution with the radar observation is derived based on the Bayesian theory. Finally simulation results show that the cued search algorithm with a known RCS fluctuant model can achieve the best performance, and the algorithm with the detection probability modeled as a beta distribution is better than that with a random selected detection probability because the model parameters can be updated by the radar observation to approach to the real value of the detection probability.

Analysis and Improvement of the Real-Time Segmented Pulse Compression-Detection Algorithm

Real-Time segmented pulse compression-detection is one of the key technologies of space-borne tracking receiver. Its implementation requires an optimized and dedicated hardware. The real-time processing places several constraints such as area occupied, power comumption, and speed. A number of segmented compression techniques have been proposed to overcome these limitations and decrease the processing latency. However, relatively high power loss in the partial field could limit their implementation in many current real-time systems. A good theoretical model was designed with intersection signal accumulation to enhance signal- noise-ratio （SNR） gain of detecting signal in the paper. From the experimental results it is known that this approach works well for pulse compression-detection, which is better suited for implementation in the high performance of current field programmable gate array （FPGA） with dedicated hardware multipliers.

Application of artificial neural network and information theory to detection of insulators

Information theory is used to obtain the information gain for each identification feature, and this gain is used as the weight factor for this feature to stress the role of effective feature, and the ART model based on artificial neural network theory is then used for identification thereby forming the detection system for poor insulators. Exper iments and calculations show this approach is correct and feasible.

The Detection of Inter-Turn Short Circuits in the Stator Windings of Sensorless Operating Induction Motors

This work proposes an alternative strategy to the use of a speed sensor in the implementation of active and reactive power based model reference adaptive system (PQ-MRAS) estimator in order to calculate the rotor and stator resistances of an induction motor (IM) and the use of these parameters for the detection of inter-turn short circuits (ITSC) faults in the stator of this motor. The rotor and stator resistance estimation part of the IM is performed by the PQ-MRAS method in which the rotor angular velocity is reconstructed from the interconnected high gain observer (IHGO). The ITSC fault detection part is done by the derivation of stator resistance estimated by the PQ-MRAS estimator. In addition to the speed sensorless detection of ITSC faults of the IM, an approach to determine the number of shorted turns based on the difference between the phase current of the healthy and faulty machine is proposed. Simulation results obtained from the MATLAB/Simulink platform have shown that the PQ-MRAS estimator using an interconnected high-gain observer gives very similar results to those using the speed sensor. The estimation errors in the cases of speed variation and load torque are almost identical. Variations in stator and rotor resistances influence the performance of the observer and lead to poor estimation of the rotor resistance. The results of ITSC fault detection using IHGO are very similar to the results in the literature using the same diagnostic approach with a speed sensor.

面向学生体质测量的数据异常可视化检测方法

单一的学生体质测量数据异常很难形成标准化判断方案,存在预测偏差准确性差、异常数据检测率低、异常数据误报率高的问题。提出面向学生体质测量的数据异常可视化检测方法,标准化处理数据,从而降低数据维度,清洗数据后预测出异常数据量最多的残缺数据。其次利用信息增益完成学生体侧数据特征的筛选以及排序,结合信息增益和正常数据点降低数据复杂度并识别出异常数据,实现学生体质测量异常数据检测。实验结果表明,所提方法的预测偏差准确性较高、异常数据检测率较高、异常数据误报率较低、检测运行时间较短。

Task-Specific Feature Selection and Detection Algorithms for IoT-Based Networks

As IoT devices become more ubiquitous, the security of IoT-based networks becomes paramount. Machine Learning-based cybersecurity enables autonomous threat detection and prevention. However, one of the challenges of applying Machine Learning-based cybersecurity in IoT devices is feature selection as most IoT devices are resource-constrained. This paper studies two feature selection algorithms: Information Gain and PSO-based, to select a minimum number of attack features, and Decision Tree and SVM are utilized for performance comparison. The consistent use of the same metrics in feature selection and detection algorithms substantially enhances the classification accuracy compared to the non-consistent use in feature selection by Information Gain (entropy) and Tree detection algorithm by classification. Furthermore, the Tree with consistent feature selection is comparable to the ensemble that provides excellent performance at the cost of computation complexity.

波束分裂模式检测下的太赫兹稀疏信道估计方案

针对现有的基于共同稀疏信道支持的理想假设的宽带信道估计方案中阵列增益严重损失的问题,提出一种波束分裂模式检测的信道估计改进方案。建立具有时延-相控模拟的混合预编码的宽带太赫兹大规模多输入多输出系统模型;通过波束分裂效应估计出信道分量的物理方向,检测出不同子载波的信道支持度,实现每个信道分量的估计。仿真结果表明,与其它信道估计方案相比,所提方案减少了导频开销,降低了计算复杂度,鲁棒性优越。

基于三稳态随机共振的BPSK信号检测算法

基于高斯白噪声下BPSK信号接收误码率较高的问题,提出了一种基于三稳态随机共振系统来降低BPSK信号误码率的方法。首先,基于随机共振的BPSK通信系模型,推导了系统的输出信噪比及信噪比增益的表达式。其次,以信噪比增益为衡量指标,探究了系统参数a、b、c和噪声强度D对共振输出的影响。研究表明,随着噪声强度的增加,输出信噪比呈现单峰结构;信噪比增益始终大于1,峰值随着系统参数b的增大而逐渐增大,但随着系统参数a和c的增大而减小。最后,在强噪声背景下,对BPSK信号的传统接收方案和所提方法的误码率进行仿真对比分析,结果证实信号在经过随机共振系统后,时域图像更为清晰,毛刺明显减少,频谱幅值是传统方法的4.613倍。

基于反相积分峰值检测法的高频信号AGC系统

针对现有非射频自动增益控制系统难以处理高频信号的问题,提出了一种反相积分峰值检测方法,并基于此方法设计了高频信号自动增益控制系统。反相积分峰值检测法是针对低频MCU设计的高频信号峰值检测方法,可以实现宽频带上的高精度峰值检测。系统由反相积分峰值检测模块对输出信号进行峰值检测,由MCU控制模块根据峰值检测结果调控可变增益放大模块的增益,实现自动增益控制。实验结果表明:该自动增益控制系统最高工作带宽可达200 MHz,相对误差在0.7%以内,相较现有基于非射频峰值检测法的自动增益控制系统,最高工作带宽在控制精度更高的前提下提升到2倍以上。

一种大动态范围数字 AGC快速控制算法

在大动态范围的复杂通信系统中,自动增益控制系统的性能起着至关重要的作用。针对传统自动增益控制系统的动态范围较小及收敛速度较慢的问题,提出了一种大动态范围数字AGC快速控制算法。该算法采用射频峰值检测以及大信号指示的方式实现对射频功率的快速检测;采用平均功率的方式对中频信号功率进行滑动平均,准确获取中频信号的平均功率;结合中频和射频增益控制策略,对接收信道的增益进行分配以及快速控制,使得射频信号经过射频单元后始终位于模数转换器(Analog to Digital Converter,ADC)的动态范围内,从而实现射频信号的稳定接收。仿真结果表明,相较于传统算法,所提算法的收敛时间由毫秒级提高到了微秒级,动态范围大幅提升。

异源在线网络话题早发现及演化特征研究

从海量新闻标题、微博等短文本异源数据及早、准确地发现舆情话题及其演化规律,以期为政府和企业监管舆情态势提供参考。以大规模异源在线社会网络数据为研究源,以文本关键词共现为建模依据点构建时变话题网络模型,将异源在线网络话题早发现及演化追踪问题转变为时变话题网络的动态社团发现与社团演化规律研究问题。进一步,提出以Louvain算法为迭代核心且以模块度增益和网络局部增量为算法量化对象的时变网络动态Louvain社团发现方法,通过与静态Louvain社团发现算法在大量计算机仿真和真实异源社会媒体数据下实验对比分析,证明所建话题网络和动态Louvain社团算法可在较少时空代价下快速、有效地实现对异源在线社会网络隐匿话题的早发现和演化规律追踪。