关键词： Parkinson's disease   machine learning   image processing  

摘要： Parkinson's disease is a neurodegenerative disorder that affects around 10 million people in the world and is slightly more prevalent in males. It is characterized by the loss of neurons in a region of the brain known as substantia nigra. The neurons of this region are responsible for synthesizing the neurotransmitter dopamine, and a decrease in the production of this substance may cause motor symptoms, a characteristic of the disease. To obtain a definitive diagnosis, the patient's medical history is analyzed and the subject submitted to a series of clinical exams. One of these exams that can take place in the clinical environment comprises asking the patient to create a series of specific drawings. Our work is based on asking the patients to draw using a software developed for this specific purpose. The drawings will then be passed through a series of image methods to reduce noises and extract the characteristics of 11 metrics of each drawing;finally, these 11 metrics will be stored. Machine learning techniques such as Optimum-Path Forest, Support Vector Machine remove, and Naive Bayes use the dataset to search and learn of the characteristics for the process of classifying individuals distributed into two classes: sick and healthy. (C) 2019 Elsevier B.V. All rights reserved.

关键词： 人工智能   识别系统   光纤检测  

摘要： 智能模式识别系统是将前期收集来的不同模式下的振动信号,利用TensorFlow神经网络架构实现机器深度学习,使其不断接触各种振动模式,从而能够识别多种情况下的振动模式,并向外界发出信号。此方法可以用于智能振动检测与故障诊断以及预警系统,从而保证设备的安全运行,预防和减少恶性事故的发生。基于人工智能的振动模式辨别系统主要实现智能振动监测与故障诊断,保证设备的安全运行,预防和减少恶性事故的发生,消除故障隐患,保障人身和设备安全,提高劳动生产率。

关键词： 光谱技术   无损检测   模式识别   损伤苹果  

摘要： 目的通过紫外/可见光谱技术结合模式识别算法,建立挤压损伤苹果的Fisher识别模型、K最近邻(KNN)识别模型和偏最小二乘判别分析(PLS-DA)识别模型。方法以挤压损伤苹果和无损苹果为研究对象,采用光谱仪采集2种苹果的光谱反射率,综合比较不同光谱预处理方法(二阶微分(SD)、标准正态变换(SNV)和多元散射校正(MSC))对各模型识别效果的影响,并利用主成分分析方法(PCA)对预处理后的光谱数据进行降维,并提取能反映损伤苹果的特征光谱。结果采用主成分分析法选择了累计贡献率超过99%的前7个主成分(P1-P7)作为特征光谱数据,有效地实现了光谱数据的降维;二阶微分对光谱反射率预处理的效果最好;3种判别模型均能满足实际要求,且SD+Fisher和SD+PLS-DA识别模型对校正集和预测集样本的总正确识别率均高达100%。结论研究结果有助于实现挤压损伤苹果的快速识别。

关键词： Scutellaria baicalensis Georgi (SBG)   fingerprint   chemical pattern recognition   classification  

摘要： Scutellaria baicalensis Georgi (SBG) is not just as a traditional herbal medicine but also a popular functional food in China and other Asian countries. A sensitive simple strategy was developed for the first time to analyze SBG from eight different geographical sources using high-performance liquid chromatography (HPLC) coupled with multivariate chemometric methods. Two unsupervised pattern recognition models, hierarchical cluster analysis (HCA) and principal components analysis (PCA), and a supervised pattern recognition model, partial least squares discriminant analysis (PLS-DA), were used to analyze the chemical compositions and physical traits of SBG. The important chemical markers baicalin, baicalein, and wogonoside were analyzed quantitatively and with PLS-DA. These methods distinguished rotten xylem (kuqin) and strip types (tiaoqin) of SBG and found that the thickness of the slice had a significant impact on the classification of SBG. Two classes of strip types were identified: one as the uncut pharmaceutical, which was sectioned with a thickness >3 mm;the other as a thin-sectioned strip type, with a thickness of <2 mm. This fingerprinting technique coupled to a chemometric analysis was used for the simultaneous quantitation of three components (chemical markers) of SBG, and greatly simplified the complicated identification of the multiple components of this plant relative to traditional methods. The strategy can clearly distinguish between kuqin and tiaoqin of SBG, and suggests that the thickness of the slice can be used as the basis for evaluation of SBG. These data provide a theoretical basis and scientific evidence for the development and utilization of SBG.

关键词： 局部放电   模式识别   变分模态分解   Wigner-Ville分布   堆栈稀疏自编码网络  

摘要： 针对目前常用的浅层模式识别方法对高维大容量样本处理困难的问题,提出一种基于变分模态分解(variational mode decomposition,VMD)的Wigner Ville分布(Wigner-Ville distribution,WVD)和堆叠稀疏自编码网络(stockedsparse auto-encoder,SSAE)的局部放电(partial discharge,PD)信号的模式识别方法。首先,以VMD算法对PD信号进行分解,对所得各分量进行时频分析得到相应的WVD;然后,以PD信号的VMD-WVD分布为输入量,利用SSAE对样本集合进行训练,自主提取内在特征。此外,将SSAE与稀疏自编码器(stackedsparseauto-encoder,SAE)的输出特征进行比较,验证了SSAE网络特征提取能力的优越性;最后,用训练好的SSAE网络完成测试样本的局部放电类型的识别。同时,以基于反向传播(backpropagation,BP)神经网络和支持向量机(support vector machine,SVM)的识别结果与该结果进行比较。结果表明,所采用的识别方法具有更高的正确识别率。

关键词： Pythagorean模糊语言相似测度   模式识别   医疗诊断   Pythagorean模糊语言集  

摘要： 文章提出了一种基于新的Pythagorean模糊语言相似测度的模式识别方法。首先定义了一种新的Pythagorean模糊语言相似测度,讨论其优良性质,接着提出基于新的Pythagorean测度的模式识别方法,最后通过医疗诊断模式识别案例验证新方法的可行性和有效性。

关键词： 护肝片   指纹图谱   HPLC   模式识别   五味子醇甲   柴胡皂苷a   柴胡皂苷c   柴胡皂苷d   五味子甲素   五味子乙素   质量评价   主成分分析   正交偏最小二乘法-判别分析  

摘要： 目的建立中成药护肝片的HPLC指纹图谱,为其质量评价提供依据。方法采用HPLC法,以五味子醇甲为参照,绘制10批护肝片原料粉样品的HPLC图谱;利用中药色谱指纹图谱相似度评价系统(2004 A)进行相似度评价,确定共有峰。采用主成分分析(PCA)和正交偏最小二乘法-判别分析(OPLS-DA)对样品进行模式识别分析。结果 10批护肝片样品的HPLC图谱有22个共有峰,经对照品进行化学指认共鉴定了其中的6个色谱峰,分别是五味子醇甲、柴胡皂苷a、柴胡皂苷c、柴胡皂苷d、五味子甲素、五味子乙素。10批供试品的相似度均大于0.97,表明该药物总体质量较为稳定;但通过PCA发现不同批次药物质量之间存在微小差异,且主要分为2类,最后进一步采用OPLS-DA筛选出了导致批次药物质量差异的2种主要成分,分别为12号峰(柴胡皂苷d)和22号峰。结论护肝片指纹图谱的构建和化学模式识别为护肝片的质量评价提供了可行的理论依据,可为生产厂家在药品生产中更高效、合理地控制药品质量提供科学依据。

关键词： 控制图模式识别   深度学习   双向长短时间记忆网络   并行计算   蒙特卡洛仿真  

摘要： 为提高制造过程质量智能控制的控制效果,提出了一种基于双向长短时间记忆网络(BidirectionalLSTM,Bi-LSTM)的控制图失控模式识别方法。文中分析了其分类的基本原理,构建了控制图模式识别模型,并通过蒙特卡洛仿真方法生成仿真数据集,进行仿真实验验证。仿真实验结果表明,Bi-LSTM用于控制图模式识别,准确率相对多层感知机(MLP)、贝叶斯分类器有了显著提升,相比支持向量机(SVM)具有效率上的明显优势,且在大样本下识别准确率更高。

关键词： 喇曼放大器   OTDR   模式识别  

摘要： 为了解决多跨段长传输系统中光纤故障定位不够准确的问题,针对融合光时域反射仪(OTDR)功能的喇曼放大器系统,提出一种定位OTDR曲线中非反射事件的方法。首先将OTDR曲线分割成事件区间和噪声区间,采用三阶差分算法定位事件区间中的突变点。然后利用霍夫变换方法确定差分变换后的检测阈值,该阈值可用于精确滤除突变点区间中的噪声,最后采用模板匹配法,将差分变换后的曲线中的事件识别归集为三角图像边缘检测问题。实验测试结果表明:该方法可进行阈值动态调整,能有效提升事件检测精度。