基于近红外光谱的粳稻种子快速鉴别方法研究

黑龙江省是我国最大的粳稻产区和商品粮生产基地。水稻种植过程中,选择合适的水稻品种是实现高产的关键环节。在农业生产中,水稻品种的选择受多方面因素影响,一般说来,同一积温带所种植的不同水稻品种在外观上差别不大,甚至没有差别,很难通过肉眼观察进行准确区分。为了快速鉴别肉眼不便区分的不同类别粳稻种子,提出了一种基于近红外光谱技术的粳稻品种快速无损鉴别方法。以黑龙江垦区大量种植的3种不同品种的粳稻种子（垦粳5号、垦粳6号和绥粳4号）作为研究对象,每个品种选取40个样本,其中30个样本做为建模集,10个样本作为预测集,扫描获取全部120个样本的近红外光谱数据。对原始光谱数据（11 520～4 000 cm-1）两端进行裁剪,选取吸光度较强的8 250～5 779cm-1范围内的光谱数据进行研究。首先建立参照模型,即直接对光谱数据建立BP模型1, 同时光谱数据经过一阶导数和Savitzky-Golay平滑预处理后建立BP模型2。模型1的分类正确率为93.3%,预测集均方根误差RMSEP=0.232 8,迭代时间t=3 882.9 s。模型2的分类正确率为100%,RMSEP=0.070 6,迭代时间t=954.5 s。比较两种模型的评价参数RMSEP发现FD+SG预处理可以提高模型的预测能力,但是由于两种模型未进行降维处理,数据量过大,模型的输入节点过多,迭代时间太长,不利于实际应用。因此利用小波变换多分辨率的特点对数据进行降维处理,采用预测集残差平方和Press值作为评价指标,在多个小波类别和参数中选取分解尺度为5的sym2（symlet2）小波对光谱数据进行压缩和降维处理,将光谱数据由601维降到21维。以小波变换结果作为神经网络输入,建立模型3,并与模型1比较,模型3的分类正确率为93.3%, RMSEP=0.225 0, 迭代时间t缩短至198.9 s,比较结果显示小波降维可以减少神经网络的输入,简化神经网络的结构,从而提高迭代速度,但对提高模型的预测能力效果不明显。上述三种模型比较结果表明,FD+SG预处理可以提高模型的预测能力,小波降维可以提高模型的迭代速度,综合上述三种模型的比较结果分析,最终建立“FD+SG+小波降维”的21输入、15个隐层、3个输出的神经网络鉴别模型4,其分类正确率达100%,RMSEP=0.029 3, 迭代时间为98.8 s,表明模型4能够完全实现对三种不同水稻品种的快速、准确、无损鉴别。因此,所提出的基于近红外光谱的小波降维和反向传播人工神经网络鉴别模型的方法完全可以用于粳稻种子的快速无损鉴别,同时也为其他农作物种子的快速鉴别提供了参考。     

天体光谱信号的连续谱归一化新方法

采用事先扣除强谱线的策略,提出小波变换与样条拟合相结合的方法进行连续谱拟合,并通过对实际光谱的试验,与较常用的连续谱拟合方法进行比较,结果表明文章提出的方法有明显的优越性。另外,针对多项式拟合等方法得到的连续谱可能会出现负值的情况,推导出有效的连续谱校正公式。     

基于谱线特征匹配的恒星光谱自动识别方法

我国正在实施的大型巡天项目(LAMOST项目),急需恒星光谱的自动识别系统。文章给出了一种基于谱线特征匹配的恒星光谱自动识别方法。该方法由以下主要步骤组成：(1) 利用小波变换的方法对观测光谱进行谱线特征提取;(2) 将提取出的特征和恒星谱线的特征模板进行相关匹配;(3) 根据相关匹配结果进行恒星光谱识别。通过对Sloan Digital Sky Survey (SDSS),Data Release Four (DR4)中的大量真实光谱数据实验表明,该方法具有对噪声鲁棒等特点,正确识别率高达96.7%。该方法可对相对定标的巡天光谱进行自动识别,符合LAMOST数据的要求,可为天文学家进行恒星和银河系的结构等研究提供帮助。     

DNA测序信号去噪分析的一种新方法

在DNA荧光测序中,噪声会影响分析的准确度和检出限。相比其他滤波方法,小波分析具有良好的时频域分辨特性。在小波去噪处理中,正确选择合适的小波基函数、去噪阈值和分解层数直接关系到信号去噪处理的质量。为了真实构建噪声模型并准确评价去噪算法的有效性,实验中通过实际系统中采集到的噪声信号叠加理想荧光信号构建DNA测序仿真信号,去噪分析的结果表明：选择sym7小波基函数、分解层数(lev＝5)与使用固定格式软阈值,有效去除了DNA测序信号的噪声;处理后,信号的信噪比提高了5倍以上。将其用于处理实际的DNA电泳荧光信号,相比基于随机噪声模型的算法,去噪后的信号更加真实可靠。     

Estimation of dimension functions of band-limited wavelets

The dimension function D_ψ of a band-limited wavelet ψ is bounded by n if its Fourier transform is supported in [−(2ⁿ⁺²/3)π,(2ⁿ⁺²/3)π]. For each and for each , 0<<δ=δ(n), we construct a wavelet ψ with supp

Full-size image

such that D_ψ>n on a set of positive measure, which proves that [−(2ⁿ⁺²/3)π,(2ⁿ⁺²/3)π] is the largest symmetric interval for estimating the dimension function by n. This construction also provides a family of (uncountably many) wavelet sets each consisting of infinite number of intervals.     

Multiresolution analysis on local fields

In this paper, we first establish the fundamental framework of multiresolution on local field in wavelet analysis. Accurately, orthonormal systems consisting of integer translations of a single function φ∈L²(K) are characterized (Theorems 1 and 2), the concept of multiresolution analysis on local field (Definition 3) and construction of corresponding wavelet vectors (Theorem 3) are given. The definition of integral periodicity of a function on local field (Definition 4) is also given. An example is finally presented.     

Normal Multiresolution Approximation of Curves

A multiresolution analysis of a curve is normal if each wavelet detail vector with respect to a certain subdivision scheme lies in the local normal direction. In this paper we study properties such as regularity, convergence, and stability of a normal multiresolution analysis. In particular, we show that these properties critically depend on the underlying subdivision scheme and that, in general, the convergence of normal multiresolution approximations equals the convergence of the underlying subdivision scheme.     

Beyond Besov Spaces,Part 2: Oscillation Spaces

We study several extensions of Besov spaces; these extensions include the oscillation spaces O_p^s,swhich take into account correlations between the positions of large wavelet coefficients through the scales and, more generally, spaces defined through the distributions of suprema of wavelet coefficients on wavelet subtrees. These spaces are independent of the particular wavelet basis chosen. Examples of applications will be taken from image processing and multifractal analysis.     

Hidden corrosion detection in aircraft aluminum structures using laser ultrasonics and wavelet transform signal analysis

Preliminary results of hidden corrosion detection in aircraft aluminum structures using a noncontact laser based ultrasonic technique are presented. A short laser pulse focused to a line spot is used as a broadband source of ultrasonic guided waves in an aluminum 2024 sample cut from an aircraft structure and prepared with artificially corroded circular areas on its back surface. The out of plane surface displacements produced by the propagating ultrasonic waves were detected with a heterodyne Mach-Zehnder interferometer. Time-frequency analysis of the signals using a continuous wavelet transform allowed the identification of the generated Lamb modes by comparison with the calculated dispersion curves. The presence of back surface corrosion was detected by noting the loss of the S(1) mode near its cutoff frequency. This method is applicable to fast scanning inspection techniques and it is particularly suited for early corrosion detection.     

Interpolation operators associated with sub-frame sets

Interpolation operators associated with wavelets sets introduced by Dai and Larson play an important role in their operator algebraic approach to wavelet theory. These operators are also related to the von Neumann subalgebras in the ``local commutant' space, which provides the parametrizations of wavelets. It is a particularly interesting question of how to construct operators which are in the local commutant but not in the commutant. Motivated by some questions about interpolation family and C*-algebras in the local commutant, we investigate the interpolation partial isometry operators induced by sub-frame sets. In particular we introduce the -congruence domain of the associated mapping between two sub-frame sets and use it to characterize these partial isometries in the local commutant. As an application, we obtain that if two wavelet sets have the same -congruence domain, then one is a multiresolution analysis (MRA) wavelet set which implies that the other is also an MRA wavelet set.