基于卷积神经网络的烟叶近红外光谱分类建模方法研究

卷积神经网络(CNN)在图像分类识别领域应用广泛,但其在近红外光谱分类中的研究还未见报道,对基于CNN的近红外光谱分类建模方法进行了研究。针对近红外光谱数据的特点,提出了一种改进的卷积神经网络建模方法,对CNN经典模型LeNet-5所做改进：①将方形矩阵卷积核改为适用于一维近红外光谱的向量卷积核;②简化网络结构,将LeNet-5结构中C5,F6及输出层改为单层感知机。同时,采用隔点采样的方法对近红外光谱降维,加快收敛速度;并对卷积核尺寸对建模结果的影响进行了研究。以我国东北、黄淮、西南三大烤烟产区的600个中部烟叶样本的近红外光谱为实验对象,建立烟叶产区分类NIR-CNN模型。该模型对训练集和测试集的判别准确率为98.2%和95%。实验结果表明,应用卷积神经网络可对近红外光谱数据准确、可靠地判别分类;烟叶产区NIR-CNN建模方法可为卷烟企业烟叶原料科学合理利用提供指导,为维护卷烟产品的质量稳定有重要意义;基于卷积神经网络的近红外光谱判别方法也可推广到其他农产品的分类应用中。     

基于拉曼光谱的复杂硫化矿快速分类方法研究

热液硫化物型脉状矿作为一类复杂硫化矿,其区域特征、成矿规律及矿物成份已有初步研究。由于成矿时期的不同,矿石中有用矿物的特征存在较大差异,导致不同矿物的性质变化较大。在选矿过程中,矿物性质的差异一定程度增加了选矿难度,减少了有用矿物回收率。因此,迫切需要一种快速、简单的对复杂硫化矿进行分类的方法,进而提高选矿指标。激光拉曼光谱技术作为一种能够分析物质结构信息的手段,已被应用于矿物的成份鉴定和结构分析。通过对大量矿物样本的激光拉曼光谱的研究,结合矿物性质深入揭示其光谱差异的原因,提出了一种基于拉曼光谱的复杂硫化矿矿源分类方法。实验结果表明：由于此类复杂硫化矿成矿时期的差异,从而造成矿物结构和性质存在较大差异。荧光主要由原矿中的脉石矿物产生,猝灭矿物中瞒石的荧光背景后可知201.62, 242.54, 288.38和309.77 cm-1处拉曼峰可以作为此类硫化矿的拉曼指纹谱。为此,基于此类硫化矿的荧光强度和代表谱峰强度与荧光背景比值可以将矿源分为三类,并利用工业试验结果进一步验证分类方法的准确性。本研究深入分析了此类复杂硫化矿的激光拉曼光谱与其矿物性质与类别之间的密切关系,提出了矿源快速分类方法,矿样无需经过复杂的化学前处理过程,对提高选矿作业效率具有重要应用价值。     

Detection of ochratoxin A contamination in stored wheat using near-infrared hyperspectral imaging

Near-infrared (NIR) hyperspectral imaging system was used to detect five concentration levels of ochratoxin A (OTA) in contaminated wheat kernels. The wheat kernels artificially inoculated with two different OTA producing Penicillium verrucosum strains, two different non-toxigenic P. verrucosum strains, and sterile control wheat kernels were subjected to NIR hyperspectral imaging. The acquired three-dimensional data were reshaped into readable two-dimensional data. Principal Component Analysis (PCA) was applied to the two dimensional data to identify the key wavelengths which had greater significance in detecting OTA contamination in wheat. Statistical and histogram features extracted at the key wavelengths were used in the linear, quadratic and Mahalanobis statistical discriminant models to differentiate between sterile control, five concentration levels of OTA contamination in wheat kernels, and five infection levels of non-OTA producing P. verrucosum inoculated wheat kernels. The classification models differentiated sterile control samples from OTA contaminated wheat kernels and non-OTA producing P. verrucosum inoculated wheat kernels with a 100% accuracy. The classification models also differentiated between five concentration levels of OTA contaminated wheat kernels and between five infection levels of non-OTA producing P. verrucosum inoculated wheat kernels with a correct classification of more than 98%. The non-OTA producing P. verrucosum inoculated wheat kernels and OTA contaminated wheat kernels subjected to hyperspectral imaging provided different spectral patterns.     

高光谱成像的油菜和杂草分类方法

利用高光谱成像技术结合化学计量学方法对油菜中的杂草进行分类识别。采用近红外高光谱技术,通过正态变量变换(SNV)、去趋势化(De-trending)、多元散射校正(MSC)、移动平均平滑法(MA)、多项式卷积平滑法(SG)、基线校正(baseline)及归一化(normalize)算法对光谱数据进行预处理,采用主成分载荷(PCA loadings)、载荷系数法(x-LW)、回归系数法(RC)、连续投影算法(SPA)分别进行特征波长提取,采用偏最小二乘判别分析(PLS-DA)、极限学习机(ELM)和支持向量机(SVM)建立分类模型。结果表明,基于De-trending 预处理,通过PCA loadings,x-loading weights及SPA特征波长提取方法,基于极限学习机ELM算法建立的模型取得了最优的分类效果,建模集和预测集的分类精度均达到100%,另引入平均分类精度的指标,发现不同试验时间下,模型分类精度变化不大,表明应用近红外高光谱成像技术对油菜和杂草进行分类是可行的。     

Twisted Poincaré Duality for some Quadratic Poisson Algebras

We exhibit a Poisson module restoring a twisted Poincaré duality between Poisson homology and cohomology for the polynomial algebra endowed with Poisson bracket arising from a uniparametrised quantum affine space. This Poisson module is obtained as the semiclassical limit of the dualising bimodule for Hochschild homology of the corresponding quantum affine space. As a corollary we compute the Poisson cohomology of R, and so retrieve a result obtained by direct methods (so completely different from ours) by Monnier. This research of the first author was supported by a Marie Curie Intra-European Fellowship within the 6th European Community Framework Programme. The second author was supported by EPSRC Grant EP/D034167/1.     

The Sixth Painlevé Equation as Similarity Reduction of $${\widehat{\mathfrak{gl}}_3}$$ Generalized Drinfel’d–Sokolov Hierarchy

Scaling symmetry of -type Drinfel’d–Sokolov hierarchy is investigated. Applying similarity reduction to the hierarchy, one can obtain the Schlesinger equation with (n + 1) regular singularities. Especially in the case of n = 3, the hierarchy contains the three-wave resonant system and the similarity reduction gives the generic case of the Painlevé VI equation. We also discuss Weyl group symmetry of the hierarchy.      

Ways in which engaging with someone else’s reasoning is productive

Classrooms which involve students in mathematical discourse are becoming ever more prominent for the simple reason that they have been shown to support student learning and affinity for content. While support for outcomes has been shown, less is known about how or why such strategies benefit students. In this paper, we report on one such finding: namely that when students engage with another’s reasoning, as necessitated by interactive conversation, it supports their own conceptual growth and change. This qualitative analysis of 10 university students provides insight into what engaging with another’s reasoning entails and suggests that higher levels of engagement support higher levels of conceptual growth. We conclude with implications for instructional practice and future research.     

Covering and radius-covering arrays: Constructions and classification

The minimum number of rows in covering arrays (equivalently, surjective codes) and radius-covering arrays (equivalently, surjective codes with a radius) has been determined precisely only in special cases. In this paper, explicit constructions for numerous best known covering arrays (upper bounds) are found by a combination of combinatorial and computational methods. For radius-covering arrays, explicit constructions from covering codes are developed. Lower bounds are improved upon using connections to orthogonal arrays, partition matrices, and covering codes, and in specific cases by computation. Consequently for some parameter sets the minimum size of a covering array is determined precisely. For some of these, a complete classification of all inequivalent covering arrays is determined, again using computational techniques. Existence tables for up to 10 columns, up to 8 symbols, and all possible strengths are presented to report the best current lower and upper bounds, and classifications of inequivalent arrays.     

Quadrature for hp-Galerkin BEM in {\hbox{\sf l\kern-.13em R}}^3

Summary. The Galerkin discretization of a Fredholm integral equation of the second kind on a closed, piecewise analytic surface is analyzed. High order, -boundary elements on grids which are geometrically graded toward the edges and vertices of the surface give exponential convergence, similar to what is known in the -Finite Element Method. A quadrature strategy is developed which gives rise to a fully discrete scheme preserving the exponential convergence of the -Boundary Element Method. The total work necessary for the consistent quadratures is shown to grow algebraically with the number of degrees of freedom. Numerical results on a curved polyhedron show exponential convergence with respect to the number of degrees of freedom as well as with respect to the CPU-time. Received April 22, 1996     

A new class of particle methods

Summary. Particle methods are numerical methods designed to solve problems in fluid mechanics and related problems in continuum mechanics. A general approach to the construction of such particle methods is presented in this article. The particles are no mass points but possess a finite extension. They can rotate in space and have a spin. The conservation of mass is automatically guaranteed by the ansatz. The forces of interaction between the particles are derived in a canonical way from the force laws of continuum mechanics and are directly based on a regularized stress tensor. In the absence of external forces and of heat sources and sinks, momentum, angular momentum, and energy are conserved as in the continuum case. Received February 17, 1995 / Revised version received December 28, 1995