基于健壮主成分分类器的无监督异常检测方法研究

入侵检测系统在训练过程中需要大量有标识的监督数据进行学习,不利于其应用和推广,经典主成分分析方法对离群数据非常敏感,进而导致分类准确性的下降。为了解决该问题,提出了一种基于健壮主成分分类器的方法,得到被离群数据干扰较少的主成分。根据主成分空间距离和数据重构误差构建异常检测模型。实验表明:该方法能够有效检测未知入侵,在检测率、误警率方面都达到较满意的结果。

增量式健壮主成分分类器的无监督异常检测方法研究

传统的入侵检测算法对数据集的研究都是针对静态训练数据的,对于动态数据却显得无能为力。但在实际应用中,入侵行为层出不穷,入侵检测系统应能对新的入侵行为进行增量学习。为了解决该问题,在前期工作的基础上,提出一种基于增量式分类器的无监督异常检测方法;实验表明:该方法在训练数据为动态情况下,能够有效检测未知入侵,在检测率、误警率方面都达到较满意的结果,并在效率上有较大提高。

我国入境旅游客源市场主成分分类研究

主成分分类与一般聚类分析相比,具有成果丰富直观,便于分析理解等特点。运用主成分分类方法研究我国入境旅游客源市场类型划分问题,根据SPSS主成分分析的最终成果和中间成果,按全国31个省级行政区主成分得分分值,划分出4个目的地类型,按13个主要客源国的公因子载荷数值,划分为4个客源地类型,并进一步分析了这些类型的形成原因,并就我国入境旅游客源市场开拓问题进行了初步分析与探讨。

FTIR-ATR指纹图谱的主成分分析-马氏距离法应用于烟用香精质量控制

在配制过程正常的条件下,分批采集150个三种正常烟用香精的傅里叶变换衰减全反射红外指纹图谱,并对原始数据进行均值中心化处理,同时对其光谱信号进行二阶导数基线校正和Karl Norris降噪处理,应用化学计量学中的主成分分析-马氏距离法分类建模,然后对烟用香精配制质量进行多元统计过程控制(MSPC),结果表明,将马氏距离上限的控制范围(UCL)设定在该类平均马氏距离的+3σ范围之内,均获得了正确的监控预报。

中国地方品种鸡的分类研究

中国鸡种资源极为丰富，中国历史上形成的许多优良地方品种鸡，对中国养鸡事业发展和世界养鸡业都曾做出过重要贡献。为充分利用我国地方鸡种资源遗传潜力和杂种优势，本文利用图论主成分分类法与系统聚类法根据20项指标研究了中国30个地方品种鸡的分类问题，分类结果比较符合实际，特别是图论主成分分类图直观清楚，生物学含意明确，对中国鸡种资源和基因库的利用有实际参考价值。

基于支持向量机的纹理图像分类算法

研究纹理图像的分类问题,纹理特征提取和分类器设计是决定分类正确率高低的关键。由于库存图像较多,且质量受到噪声影响,使图像特征提取比较困难。针对传统特征提取和分类算法分类正确率不高的难题,提出一种基于支持向量机的纹理图像分类算法。首先采用Gabor滤波器对纹理特征进行提取,采用主成分分析对提取后的特征进行选择,最后采用支持向量机进行纹理图像的分类。采用Brodatz纹理库进行测试实验,实验结果表明,支持向量机分类算法提高了纹理图像的分类正确率,降低了误分率和拒分率,且分类速度加快,适用于更为复杂的纹理分类,为图像提取提供了参考。

全民健身时代下运动类软件使用影响因素研究——以Keep软件为例

本文以Keep软件为例,先是通过基本描述性统计分析对Keep软件的使用人群以及不同人群的各种使用情况进行研究,以实现对Keep软件功能的分析;然后运用分类数据主成分分析方法挖掘出潜在客户。最终得出以下结论:第一,Keep软件具有很大的使用人群,但人们使用频率普遍较低且使用的功能较基础;第二,年龄对是否使用Keep软件有显著影响,用户整体基本呈现年轻化;第三,不同年龄段使用软件的原因各不相同,可以此为依据挖掘潜在客户。最后,针对分析结论提出对应建议。

基于光谱融合的火星表面相关矿物分类方法研究

多源数据融合能在一定程度上扩展数据信息量,更利于建立准确和稳健的分析模型。行星探测中常采用多个载荷协同分析同一目标,因此利用多载荷数据融合辨别分析火星矿物具有重要科学意义和应用前景。分别采用可见近红外(Vis-NIR)反射光谱和拉曼(Raman)散射光谱两种技术手段测量了火星表面主要矿物(硅酸盐、硫酸盐、碳酸盐)的光谱特征曲线,并对获取的光谱数据进行基线校正、Savitzky-Golay平滑以及标准矢量归一化(SNV)等必要的数据预处理。根据光谱特征,首先选取样品Vis-NIR和Raman数据信息丰富、信噪比高、光谱信号重叠小的波段(Vis-NIR:430~2 430nm,Raman:130~1 100cm^(-1)),然后运用软独立建模分类法(SIMCA)、主成分分析法-K最邻近分类法(PCA-KNN)分别建立基于Vis-NIR,Raman及两者融合(累加融合、串联融合)的矿物聚类分析模型。采用SIMCA算法的矿物聚类准确率由单一光谱建模的72.6%(Vis-NIR),90.7%(Raman)提升为融合建模的96.3%(累加融合)和98.1%(串联融合);采用PCA-KNN的准确率由单一光谱建模的68.9%(Vis-NIR),72.9%(Raman)提升为融合后的80.3%(累加融合)和92.6%(串联融合)。实验结果表明:光谱融合能够发挥Vis-NIR,Raman各自的数据优势,所建火星表面相关矿物分类模型的预测准确度更高。该研究为我国火星探测任务奠定了岩石分类方法基础。

中职学生主观幸福感影响因素研究

本文通过问卷调查,从家庭、学校、学业、友谊、自由和环境六个维度来研究中职学生主观幸福感的状况,从个体、家庭、学校和社会四个方面寻求影响主观幸福感的因素,利用主成分改进的多项Logistic回归模型研究各因素的影响程度。

基于CATPCA的优化Transformer卫星电源消耗时序预测研究

提出一种由基于最优尺度量化的分类主成分分析数据处理模块和优化Transformer时序预测模块组成的卫星电源消耗预测方法.针对卫星工程数据的高冗余问题,建立了基于赫斯特指数分析(Hurst)、灰色关联分析以及分类主成分分析(CATPCA)的卫星高维数据处理模型,对百维度时序数据进行有效提取,重构输入数据.采用对抗学习网络架构,建立多学习Transformer的卫星电量预测模型,模型综合考虑影响卫星能源消耗的多种因素以及时序数据依赖,可以在较短的时间内完成高精度卫星电源消耗时序预测.实验部分采用卫星真实运行数据,综合考虑影响卫星能源消耗的多种因素,12 h预测拟合优度达到94%,比BP神经网络,长短期记忆网络(LSTM)精度更高.可以有效克服常规工程数据的冗余、缺失以及脏数据问题,解决了常规时序预测需要依赖长期数据的不足缺陷,有效完成卫星能源短时消耗高精度预测.这对卫星在轨任务规划、卫星在轨健康管理等后续任务提供可靠支持.

Study on Phenological Character in Seedling Period of Introduced Populus deltoids Clones

In order to provide directionally genetically improved breeding materials of poplar by exploring the phenological traits genetic variation level and its develop- ment potential of Populus deltoides and the resource evaluation was carried out; 8 phenological phases in seedling period were observed and analyzed of 60 Populus deltoids clones introduced from America. The results showed that： （1） there was obvious difference in phonological character among clones, especially in leaf-spread- ing peak stage and the end term of leaf-falling stage, with the largest variation co- efficient of 14.97% and the minimum of 3.83% respectively. （2） Leaf-spreading peak stage scattered but the end term of leaf-falling stage concentrated the most. The phonological character in early stage of seedling growth was the main factor influ- encing the length of growing season. （3） By principal component analysis, pheno- logical phases were classified into 3 typical periods, including germination stage, leaf-spreading peak stage and leaf-falling stage. （4） Totaling 60 clones were classi- fied into 4 types by using clustering analysis in phenological time variables of clones.

Application of PCA and HCA to the Structure-Activity Relationship Study of Fluoroquinolones

Density functional theory （DFT） was used to calculate molecular descriptors （properties） for 12 fluoro-quinolone with anti-S.pneumoniae activity. Principal component analysis （PCA） and hierarchical cluster analysis （HCA） were employed to reduce dimensionality and investigate in which variables should be more effective for classifying fluoroquinolones according to their degree of an-S.pneumoniae activity. The PCA results showed that the variables ELUMO, Q3, Q5, QA, logP, MR, VOL and △EHL of these compounds were responsible for the anti-S.pneumoniae activity. The HCA results were similar to those obtained with PCA.The methodologies of PCA and HCA provide a reliable rule for classifying new fluoroquinolones with antiS.pneumoniae activity. By using the chemometric results, 6 synthetic compounds were analyzed through the PCA and HCA and two of them are proposed as active molecules with anti-S.pneumoniae, which is consistent with the results of clinic experiments.

滨白鸡纯系及杂交组合的种群关系分析

本文采用中国科学院遗传研究所研制的Ａ、Ｂ、Ｃ三个系统１２个血型因子，分析滨白鸡七个纯系和五个杂交组合的种群关系。结果表明，１２个群体可分为五个类。第一类包括５个纯系（ＮＯ２、ＮＯ７、ＮＯ６、ＮＯ３、ＮＯ８）和２个杂交组合（５×２、７×８）．第二类包括５×６组合和ＮＯ５系。第三类是４×２组合。第四类是５×４组合。第五类是ＮＯ４系。不同类品系间遗传差异大，同一类品系间则小。实际配合力测定得到的４×２、５×４２、６×４２优秀杂交组合均有一定的理论基础。

Study on Rhizome Crops by Fourier Transform Infrared Spectroscopy Combined with Wavelet Analysis

In order to distinguish 8 kinds of rhizome crops, the 40 samples were studied by Fourier transform infrared spectroscopy （FTIR） combined with wavelet transform （WT）, principal component analysis （PCA） and hieramhical cluster analysis （HCA）. The results showed that the infrared spectra were similar on the whole, but there were differences in peak position, peak shape and peak absorption intensity in the range of 1 800-700 cm-1. The infrared spectra in the range of 1 800-700 cm-1 were selected to perform continuous wavelet transform （CWT） and discrete wavelet transform （DWT）. The 15th-Ievel decomposition coefficients of CWT and the 5=-level detail coefficients of DWT were classified by PCA and HCA. The cumulative contri- bution rates of the first three principal components of CWT and DWT were 93.12% and 89.78%, respectively. The accurate recognition rates of PCA and HCA were all 100%. It is proved that FTIR combined with WT can be used to distinguish different kinds of rhizome crops.

Morphological Diversity Analysis of Red-seed Watermelon (Citrullus lanatus ssp. vulgaris var. megalaspermus Lin et Chao) Germplasm Resources

[Objective] This study aimed to analyze the morphological diversity of red- seed watermelon （Citrullus lanatus ssp. vulgaris var. megalaspermus Lin et Chao） germplasm resources. [Method] Multiple cluster analysis and principal components analysis on the morphological traits of 51 red-seed watermelon germplasm resources were carried out. [Result] The coefficient of variations （CVs） of 39 morphological traits in 51 red-seed watermelon idioplasm resources ranged from 5.37% to 66.95%, with an average of 22.87%. The average of Shannon diversity information indices was 1.55. Among them, the Shannon diversity information index of seed length was the highest （2.16） and that of seed shell figure pattern was the lowest （0.32）. In ad- dition, the morphological diversity information indices of quantity characters were higher than that of quality characters. The principal components analysis revealed that the variance contribution rates of the first, second and third principal compo- nents were 19.49%, 15.32% and 9.55%, respectively. Cluster analysis divided the 51 materials into three broad branches based on the morphological traits. There was only one material in the fist branch and two in the second branch, and all the three materials were wild. The other 48 materials were divided into the third branch and all of them were cultivars. [Conclusion] This study provided a theoretical basis for the protection and utilization of red-seed watermelon resources.

VR背景下网店与实体店消费选择影响因素研究

随着大数据挖掘技术的高速发展,VR技术不断创新,实体店应该怎样应对网店带来的冲击成为当今零售业关注的热点。本文通过对各年龄层次的消费者进行实证调研,研究了影响实体店购买倾向的因素及VR技术的出现对消费者在实体店购物倾向的影响。首先,运用排序Logit方法研究了消费者愿意在实体店购买的因素,结果表明年龄、上网时间和网上购物频率对于用户在实体店购买的意愿有显著影响,而从事的工作行业、性别、最高学历及月消费都对在实体店的购物倾向没有显著性影响。然后,运用分类主成分分析研究了消费者不愿意在实体店购买的关键在价格优惠力度、购物时间、购物效率、产品与服务的丰富性四方面。最后,运用独立样本T检验与方差分析等方法研究了VR对消费者选择的影响。

Study on Trace Elements in Rehmannia glutinosa Libosch. by Principal Component Analysis and Clustering Analysis

[Objective] This study aimed to investigate the trace elements in Rehman- nia glutinosa Libosch. by using principal component analysis and clustering analysis. [Method] Principal component analysis and clustering analysis of R. glutinosa medicinal materials from different sources were conducted with contents of six trace elements as indices. [Result] The principal component analysis could comprehen- sively evaluate the quality of R. glutinosa samples with objective results which was consistent with the results of clustering analysis. [Conclusion] Principal component analysis and clustering analysis methods can be used for the quality evaluation of Chinese medicinal materials with multiple indices.

Comparative Analysis of Morphologic Traits of 50 Large-flowered Chrysanthemum Varieties

With 50 large-flowered Chrysanthemum varieties from germplasm nursery of Wunaoshan Forest Farm in Macheng City as research objects, 64 morphological traits were investigated by field experiments adopting randomized block design. The morphological differences were observed by uniformity analysis, variability analysis, principal component analysis and cluster analysis. The result showed that the vari- able coefficients of 59 traits were greater than 15%; the contribution rate of first seven principal components reached 81.45%; and it was found by clustering analy- sis that the 50 germplasm resources could be divided into four clusters with obvious morphological differences, and plant type could be used as an index for classifica- tion.

CLAHE低频DCT系数重变换在人脸识别中的应用

针对光照变化显著影响自动人脸识别系统性能的问题,为了规范化光照变化以提高人脸识别率,提出基于对比度限制自适应直方图均衡化CLAHE(Contrast Limiting Adaptive Histogram Equalization)的低频离散余弦变换DCT(Discrete Cosine Transform)系数重变换算法。首先将图像划分成多个互不重叠的局部小块,使用CLAHE对局部小块进行局部对比拉伸以实现去噪;然后,通过缩减适当数目的低频DCT系数来消除人脸图像中的光照变化;最后,利用核主成分分析进行特征提取,稀疏系数重建和k-近邻分类器完成最终的人脸识别。在扩展Yale B及AR人脸数据库上的实验验证了算法的有效性。实验结果表明,算法在识别非常困难的Yale B子集5上的识别率可高达98.20%,相比其他几种规范化技术,该算法取得了更高的识别率,同时大大降低了识别所耗时间。