高光谱图像技术检测柑橘果锈

高光谱图像技术作为农产品无损检测的新技术,探讨了其在柑橘外部品质检测的可行性.以检测柑橘果锈为目的,首先对经预处理的高光谱图像数据进行主成分分析,优选出571 nm、652 nm和741 nm三个特征波长组成新的图像块;再进行第二步主成分分析,得到的第三主成分图像为最适宜检测柑橘果锈的图像;最后对该图像进行中值滤波、平方根变换,阈值分割和数字形态学运算完成特征提取.试验结果表明,此算法对柑橘果锈检测的正确率可达到90%.研究表明,利用高光谱图像技术结合两步主成分分析算法检测柑橘果锈是可行的.     

紫外/可见光谱技术无损检测苹果的挤压损伤

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。     

消光法粒度测量中折射率反演问题研究

多波长消光法粒度测量中,在已知粒度分布前提下,采用遗传算法可以对折射率进行反演.相比于传统的优化方法,该算法获得的折射率结果逼近全局最优解.反演采用的波长数目和颗粒单分散性的优劣影响解的可靠性.     

基于紫外/可见光谱技术的李子硬度无损检测

目的探究采用紫外/可见光谱技术结合化学计量学预测李子硬度的可行性。方法以“红”李子和“青”李子为研究对象,采用光谱采集系统获取李子样本的平均光谱;采用标准正态变换对原始光谱数据进行预处理,并利用连续投影算法(SPA)和竞争性自适应重加权算法(CARS)从全光谱的1024个波长中分别提取2个(513.04 nm和636.72 nm)和10个(230.01,244.67,274.71,287.66,290.90,300.59,311.78,423.08,515.39,631.31 nm)特征波长;分别建立基于全光谱和提取的特征波长预测李子硬度的误差反向传播(BP)网络模型。结果将采用SPA和CARS特征波长选择方法提取的特征变量作为BP网络输入,明显提升了BP网络模型的运行效率,且SPA-BP网络模型具有相对较好的李子硬度预测能力(rp=0.695,预测样本集均方根误差为1.610 kg/cm2)。结论采用紫外/可见光谱技术结合特征波长提取方法可实现李子硬度的快速无损检测。     

基于聚类分析的光谱重建样本分析

目的 解决目前光谱重建中因数量大而出现的冗余和繁重等问题,证明聚类算法可以很好地应用在光谱选择样本分析中,并可以实现较高的重构色度精度和物理精度。方法 采用主成分分析法进行仿真实验,首先探究主成分个数,再确定聚类个数,然后比较聚类方法和3种常用的样本选择方法,最后分析比较光源种类对重构结果的影响。结果 通过实验确定主成分个数为6且聚类个数为20时,在A光源下使用KFCM算法的重构效果最好,此时平均色差为0.35ΔE00,平均RMSE为0.0078,平均GFC为99.94%。结论 聚类方法可以应用于光谱成像过程中训练样本选择过程,且有助于提高光谱重构的运算速度和精度。     

基于紫外可见吸收光谱的水质检测算法研究

为实时有效地检测地表水中硝酸根离子和亚硝酸根离子的变化过程,提出一种基于紫外可见吸收光谱的水质检测算法。针对水质光谱数据受到干扰易出现波动误差的问题,采用小波变换对其进行分解以滤除高频噪声,并通过主成分分析对数据特征进行降维以防止模型复杂度较高导致过拟合。水质光谱数据经预处理后采用支持向量机对其进行建模,通过非线性自适应调整变异收缩因子对差分进化算法进行改进,并采用改进差分进化算法对水质预测模型进行参数优化。通过与采用其他常用算法所建模型进行对比分析,实验结果表明:基于该算法所建的硝酸根离子和亚硝酸根离子模型具有更高的预测精度,且其能够以更快的收敛速度使模型达到全局最优。     

基于波长分区的半色调色光谱估算模型

目的研究半色调打印色的光谱呈色特性,为半色调打印色的光谱复制提供估算模型。方法分析不同阶调值的半色调打印色的光谱反射曲线特性,根据半色调打印色的光谱微分找出特征波段,并进行波长分区。在不同的波长分区分别使用网点增大补偿与多项式算法估算半色调打印机的光谱反射率。结果提出的半色调光谱估算算法估算的光谱与实际测量的s RMS最大值为0.02,对应的估算色与测量色的平均色差小于1.5ΔE76。结论使用波长分区的光谱估算模型可提高半色调复制色的光谱估算精度,半色调彩色输出设备的光谱特征化可以使用文中的估算模型作为正向特征化模型。     

消光法粒度测量中折射率反演问题研究

利用多波长消光法,在已知粒度分布前提下,采用遗传算法对折射率进行了反演。相比于传统的优化方法通常获得的局部最优解,该算法获得的折射率结果逼近全局最优解。反演采用的波长数目和颗粒单分散性的优劣影响解的可靠性。尽量多地获得透射谱信息以及增大颗粒系的分散性可以是测量结果更准确。由于亚微米和纳米级颗粒的消光系数-折射率曲线的单值性关系比微米级颗粒更为明显,导致其目标函数的多峰分布程度得以降低,使反演的抗噪能力增强,折射率更易于获得。     

遗传算法用于颗粒粒度分布重建的研究

消光法是一种基于消光定律的非接触式颗粒浓度和颗粒粒度测量的有效方法。由于它原理简单、易操作、易实现，因而得到了广泛的应用与研究。粒度谱重建算法是消光法粒度谱测量的核心内容，为了克服以往粒度谱重建算法需要先验知识的局限性，提出了采用改进的遗传算法来重建颗粒的粒度分布函数，并给出了算法的实现步骤。从数值模拟和实验研究的结果来看，该算法具有较高的精度和稳定性，它不受颗粒粒度分布影响，对噪声不敏感，同时还可以用于未知光学常数颗粒的测量。     

三维脑部电阻抗成像技术研究

电阻抗成像(electric impedance tomography,EIT)是利用测量得到边界电压重构内部电导率分布的技术.通过在ANSYS中建立三维头球模型并注入电流,采用自适应局部增强微分进化算法进行逆问题重构.首先对EIT的基本原理进行阐述,其次介绍微分进化算法在电阻抗成像中的应用,最后通过在头球模型上进行仿真实验,针对基本微分进化算法及其改进算法进行逆问题重构.仿真结果表明,改进的算法较微分进化算法在时间上有显著提高,适应性与牛顿-拉斐逊算法相比明显提高.     

Spherical particle size determination by analytical inversion of the UV-visible-NIR extinction spectrum

An analytic inversion method, based on the anomalous diffraction approximation for nonabsorbing spherical particles, was developed to retrieve the size distribution from the optical turbidity or extinction spectrum. This method makes use of a differential Fourier cosine transform approach and provides a simple and fast inversion by means of fast Fourier transform and the Savitzky-Golay filter. The applicability of this algorithm was tested on the extinction data generated by the Mie solution. The effects of noise, modality, band limits, and data set size were analyzed by comparison with simulated data. This method can be used to reconstruct the original monomodal and bimodal distributions from 10% noise-corrupted data. The peak position and ratio of peak heights can be recovered with 10% or less deviation. The experiments with latex spheres showed that the inversion result from this method compares favorably with that from the dynamic light scattering measurement.     

Microphysical particle parameters from extinction and backscatter lidar data by inversion with regularization: theory

A method is proposed that permits one to retrieve physical parameters of tropospheric particle size distributions, e.g., effective radius, volume, surface-area, and number concentrations, as well as the mean complex refractive index on a routine basis from backscatter and extinction coefficients at multiple wavelengths. The optical data in terms of vertical profiles are derived from multiple-wavelength lidar measurements at 355, 400, 532, 710, 800, and 1064 nm for backscatter data and 355 and 532 nm for extinction data. The algorithm is based on the concept of inversion with regularization. Regularization is performed by generalized cross-validation. This method does not require knowledge of the shape of the particle size distribution and can handle measurement errors of the order of 20%. It is shown that at least two extinction data are necessary to retrieve the particle parameters to an acceptable accuracy. Simulations with monomodal and bimodal logarithmic-normal size distributions show that it is possible to derive effective radius, volume, and surface-area concentrations to an accuracy of +/-50%, the real part of the complex refractive index to +/-0.05, and the imaginary part to +/-50%. Number concentrations may have errors larger than +/-50%.     

Microphysical particle parameters from extinction and backscatter lidar data by inversion with regularization: simulation

A sensitivity study with an inversion scheme that permits one to retrieve physical parameters of tropospheric particle size distributions, e.g., effective radius, volume, surface-area, and number concentrations, as well as the mean complex refractive index from backscatter and extinction coefficients at multiple wavelengths is presented. The optical data for the analysis are derived from Mie-scattering calculations for monomodal and bimodal logarithmic-normal distributions in the particle size range between 0.01 and 10 microm. The complex refractive index is taken between 1.33 and 1.8 in the real part and between 0 and 0.1 in the imaginary part. The choice of these parameters takes account of properties of optically active atmospheric particles. The wavelengths were chosen at 355, 400, 532, 710, 800, and 1064 nm for the backscatter and at 355 and 532 nm for the extinction data, which are the available wavelengths of the two lidar systems at the Institute for Tropospheric Research. Cases of erroneous optical data of the order of as much as 20%, an unknown refractive index, which may also be wavelength and size dependent, as well as the a priori unknown modality of the particle size distribution were considered. It is shown that both extinction channels are necessary for determining the above-mentioned parameters within reasonable limits, i.e., effective radius, surface-area, and volume concentrations to an accuracy of +/-50%, the real part of the complex refractive index to +/-0.1, and the imaginary part to +/-50%. The number concentration may have errors larger than 50%. The overall performance of the inversion scheme permits the evaluation of experimental data on a routine basis.     

Extension of the graphical technique for estimation of particle size distribution parameters for the consistent intercomparison of diverse sets of multiwavelength lidar derived optical coefficients

In applying the graphical technique to the estimation of the particle size distribution (PSD) parameters, determination of proper bounds surrounding the solution space for a particular confidence level is essential to the consistent intercomparison of diverse multiwavelength lidar optical data sets. The graphical technique utilizes ratios of backscatter and/or extinction coefficients, and it is shown that if the correlation between ratios is not taken into account in calculating the error bounds, the solution space will be overestimated, resulting in relatively larger discrepancies for a larger number of optical coefficients. A method for correcting the bounds, to account for the correlation is developed for various numbers of wavelengths. These improved bounds are then applied, for the case of a monomodal lognormal PSD, with an assumed refractive index, to assess the role additional Raman extinction channels play in improving retrieval capability of a typical three-channel backscatter lidar (1064, 532, and 355 nm) under varying noise levels. Applying the same formalism to underlying bimodal distributions of coarse and fine particles can result in false monomodal solutions. However, when both Raman optical extinction channels are available, no solution is obtained. This can potentially serve as a quick and simple method, prior to a more complex regularization analysis, to differentiate between cases in which the fine mode is dominant versus the cases in which the contribution from the coarse mode is significant.     

基于左矩阵分式模型的短记录数据模态参数识别

在测量数据有限情况下,由于难以获得频响函数(FRF)的准确估计,使用FRF 作为原始数据的传统模态参数识别方法将不再适用。针对该问题,提出一种基于频响函数左矩阵分式模型的模态参数识别方法。该方法直接使用输入输出数据FFT 谱(IO 谱)作为原始数据,避免了频响函数估计。通过最小二乘估计在Z 域内求解模态参数,改善了矩阵的求解性态。针对左矩阵分式模型的特点,给出了一种通过主分量分析(PCA)建立稳定图的方法。最后采用GARTEUR 飞机模型建立仿真算例对所提出的方法进行了验证。     

基于改进R-R分布的现场激光粒度仪反演算法

针对现场激光粒度仪中颗粒粒径分布的反演计算方法,对非独立模式法中的R-R分布法进行了研究.同时设定多组R-R分布的尺寸参数X和分布参数M,采用多个线程同时反演计算目标函数(拟合误差).经过最后比较,找出函数值最小的目标函数,即认为现场颗粒的粒径分布服从该目标函数对应的R-R分布.现场激光粒度仪的实验结果表明,在使用5个线程时,每反演一次的平均时间小于1 min.该改进方法的原理简单,准确度和重复性高,计算速度快,能够满足现场颗粒粒径的实时测量要求.     

Application of Vis–NIR hyperspectral imaging in classification between fresh and frozen-thawed pork Longissimus Dorsi muscles

Fresh and frozen-thawed (F-T) pork meats were classified by Vis–NIR hyperspectral imaging. Eight optimal wavelengths (624, 673, 460, 588, 583, 448, 552 and 609 nm) were selected by successive projections algorithm (SPA). The first three principal components (PCs) obtained by principal component analysis (PCA) accounted for over 99.98% of variance. Gray-level-gradient co-occurrence matrix (GLGCM) was applied to extract 45 textural features from the PC images. The correct classification rate (CCR) was employed to evaluate the performance of the partial least squares-discriminate analysis (PLS-DA) models, by using (A) the reflected spectra at full wavelengths and (B) those at the optimal wavelengths, (C) the extracted textures based on the PC images, and (D) the fused variables combining spectra at the optimal wavelengths and textures. The results showed that the best CCR of 97.73% was achieved by applying (D), confirming the high potential of textures for fresh and F-T meat discrimination.     

基于滑动中值滤波的多尺度主元分析方法

提出了一种基于滑动中值滤波的多尺度主元分析(MSPCA)方法,该方法利用中值滤波对主元分析(PCA)前的原始数据进行预处理,以去除异常点,并用多尺度主元分析方法把小波变换和主元分析有机结合起来,通过对过程数据的多尺度建模,来消除系统中的次要主元和小的小波系数,这样既提高了对数据中细微、重要变化的检测灵敏度,又解决了在测量数据中含有异常点的情况下,现有多尺度主元分析难以去除因异常点的存在而产生的虚警问题.仿真验证了该方法的有效性和可行性.     

Differentiation of algae clones on the basis of resonance Raman spectra excited by visible light

Fourteen algae clones belonging to four different classes, including clones of Pseudo-nitzschia (Bacillariophyceae), some of which are capable of producing the toxin domoic acid, have been studied by means of resonance Raman spectra excited at 457.9 and 488 nm. Spectra taken at both excitation wavelengths are of high quality and are sufficiently distinct to differentiate clones at the algal class level. All spectra contain major features near 1000, 1153, and 1523 cm(-)(1), which are strongly resonance enhanced due to carotenoid pigments. Weaker features between 920-980 and 1170-1230 cm(-)(1), also due to carotenoid pigments, are more characteristic of the algae clones and more directly reflect different carotenoid composition. Similarities and differences among spectra have been analyzed by the method of principal component analysis (PCA). A distinct clustering of spectral data according to algal class has been shown by PCA score plots. All Pseudo-nitzschia clones can be separated from other classes of algae on the basis of spectra, but it is not possible to distinguish toxic Pseudo-nitzschia from nontoxic clones on the basis of these spectra, which reflect only differences in carotenoid composition.     

猕猴桃挤压损伤高光谱快速检测研究

目的 探究猕猴桃挤压损伤较优的快速无损判别方法。方法 利用高光谱成像系统获得所有猕猴桃的高光谱图像,并提取猕猴桃损伤区域以及完好无损区域的光谱反射率;运用多元散射校正方法对原始反射光谱进行预处理,并运用主成分分析对光谱数据降维;比较并分析Fisher判别分析方法以及简化的K最近邻(Simplified K Nearest Neighbor,SKNN)模式识别方法对猕猴桃挤压损伤的判别效果。结果 在710~850 nm和960~1 030 nm这2个波段内,猕猴桃损伤区域的平均光谱反射率与完好无损区域的平均光谱反射率存在较明显差异;采用主成分分析从256个全波段中筛选了前5个主成分作为新变量,识别模型的检测效率得到了提升;构建的SKNN和Fisher模型对预测集中样本的正确识别率均为93.3%,从SKNN识别模型的混淆矩阵中得出,预测集中仅有2个样本出现误判,并且SKNN模型对校正集中样本的正确识别率高于Fisher模型。结论 在判别猕猴桃挤压损伤时,SKNN识别模型具有相对较好的判别效果。