结合LLE流形学习和支持向量机的猪肉颜色分级

猪肉颜色分级是肉品品质无损检测的重要内容。该文通过对猪肉彩色数字图像颜色规律的分析,结合猪肉红(R)、绿(G)、蓝(B)颜色分量及其亮度和饱和度值,构建了一个新的颜色维度。将采集的1070份猪肉图像样本进行专家分级和标记后,给出了猪肉颜色的特征表示,并利用支持向量机(SVM)进行了颜色分级。试验结果表明,随着训练集样本数的增加,分级准确率有所提高。当将所有样本作为训练集时,分级准确率高达96.5%。针对SVM分级后泛化能力不强的问题,采用流形学习LLE维数变换,使其泛化能力由37%提高近1倍。结果表明LLE可有效改善SVM的分级准确率。该方法可为猪肉品质无损检测的研究与应用提供参考。     

基于局部判别映射算法的玉米病害识别方法

如何快速准确检测到作物病害信息是作物病害防治中的一个首要问题,根据作物叶片症状识别作物病害是作物病害检测的一个基本方法。由于病害叶片颜色、形状和纹理之间的差异很大,使得很多经典的模式识别方法不能有效地应用于作物病害识别中,为此提出了一种基于局部判别映射(local discriminant projects,LDP)的作物病害识别方法。首先,利用区域增长分割算法分割病害叶片中的病斑图像;然后,将病斑图像重组为一维向量,再由LDP对一维向量进行维数约简;最后,利用最近邻分类器识别作物病害类别。利用LDP算法将高维空间的一维向量样本点映射到低维子空间时,能够使得类内样本点更加紧凑,而类间样本点更加分离,从而得到最佳的低维分类特征。利用该方法在5种常见玉米病害叶片图像数据库上进行了病害识别试验,识别精度高达94.4%。与其他作物病害识别方法(如基于神经网络、主分量分析+概率神经网络和贝叶斯方法)和监督子空间学习算法(如算法局部判别嵌入和判别邻域嵌入)进行了比较。试验结果表明,该方法对作物病害叶片图像识别是有效可行的,为实现基于叶片图像处理技术的作物病害的田间实时在线检测奠定了基础。     

苹果采摘机器人夜间图像降噪算法

苹果采摘机器人图像处理系统采集到的实时夜间图像含有大量的噪声,影响采摘效率。通过差影法对夜间图像进行噪声分析,判定其噪声类型为以高斯噪声为主,并伴有部分椒盐噪声的混合噪声。针对高斯噪声去除难题,将独立成分分析(independent component analysis,ICA)理论引入夜间图像降噪,并尝试采用粒子群优化算法(particle swarm optimization,PSO)对ICA进行优化,建立基于PSO优化的ICA降噪算法(PSO-ICA),以期最大限度地降低夜间图像的噪声污染。利用标准Lenna图像和自然光下的苹果图像,进行仿真试验,结果表明PSO-ICA方法降噪效果最为理想。然后对白炽灯、荧光灯、LED灯3种不同的人工光源下采集到10个样本点的夜间图像进行验证试验,结果表明,从视觉效果评价,在3种人工光源环境下,PSO-ICA降噪方法得到低噪图像均表现为噪点明显减少;从相对峰值信噪比(relative peak signal-to-noise ratio,RPSNR)看,在3种人工光源下的平均值,PSO-ICA得到的低噪图像,分别比原始图像、均值滤波降噪和ICA降噪得到的图像的相对峰值信噪比提高21.28%、12.41%、5.53%;从运行时间看,PSO-ICA方法较ICA方法的运行时间平均减少了49.60%。PSO-ICA方法用于夜间图像降噪有着独到的优势,为实现苹果采摘机器人的夜间作业打下坚实的基础。     

基于噪声类型及强度估计的狭叶锦鸡儿叶切片图像盲去噪

狭叶锦鸡儿叶切片显微图像在获取过程中不可避免的受到噪声污染,会对后续处理造成不良影响。针对现有噪声类型未知,去噪算法存在速度慢、效果不理想等问题,该文提出图像噪声类型估计-强度估计-去噪这一处理过程,实现对狭叶锦鸡儿叶切片显微图像降噪目的。首先采用平滑区直方图重构和拟合法确定噪声类型;然后在此基础上,应用基于图像块的SVD(singular valuable decomposition,SVD)域图像噪声强度估计法对噪声标准差进行估计;最后在确定噪声类型和强度基础上,采用几何均值滤波(geometric mean filtering,GMF)和三维块匹配滤波(block-matching and 3-D filtering,BM3D)对图像进行联合去噪。试验结果表明:该文噪声类型估计法估计出切片图像噪声类型为加性高斯噪声,高斯函数对随机选取的15幅狭叶锦鸡儿叶切片图像平滑区域直方图数据点拟合优度2R均值为0.996,平均均方根误差RMSE(root mean squared error,RMSE)为0.144 6;采用该文噪声强度估计法估计出的切片图像噪声标准差???[2.5,4.0],处理标准差较小噪声,该文算法处理精度、运行速度和稳定性等方面存在明显优势;GMF-BM3D算法在较好去除图像噪声同时,极大的保留了图像纹理、边缘和细节等信息,同时极大的提高了算法运行速度,处理后的图像BRISQUE(blind/referenceless image spatial quality evaluator,BRISQUE)值为10左右,相当于原图BRISQUE值的1/2左右。与传统BM3D算法相比,去噪效果相当,但耗时约相当于传统BM3D算法的1/9。与小波去噪算法(wavelet threshold,WT)算法相比,虽速度相对较慢,但去噪后图像BRISQUE值比使用WT法低4左右。因此,该算法较好实现了对狭叶锦鸡儿叶切片图像准确降噪,为其后续处理提供了可靠技术支持。     

基于对比度受限直方图均衡化的水下海参图像增强方法

针对水下图像受到水下复杂光照的影响导致图像对比度差的现象,采用对比度受限自适应直方图均衡化方法(contrast-limited adaptive histogram equalization,CLAHE)对水下海参图像进行增强处理,算法首先将原始图像分割成若干个子区域并且大小相同,再选取特定值对每个子区域的直方图进行截取,并将截取下的像素均匀分配到每个灰度级,最终得到限定对比度直方图。并通过研究算法中的相关参数,得到适用于水下海参图像增强的参数值,取得了更好的增强效果。通过评价函数均方差(mean squared error,MSE),峰值信噪比(peak signal to noise rate,PSNR)和信息熵(information entropy)对比CLAHE方法和其他一些方法,结果显示CLAHE算法在水下海参图像提高质量和保持图像细节方面表现出更好的性能,为以后水下机器人的识别定位提供了方便。     

像素自相关矩阵的阈值自适应角点检测算法

针对Harris角点检测算法中角点响应函数(corner response function,CRF)系数阈值与非极大值抑制系数阈值需要人为设定所造成的可变性和随机性等问题,该文提出一种通过计算图像每个像素的自相关矩阵行列式值,构造特征角点图像进行自适应阈值分割的改进Harris角点检测算法.该算法首先通过计算原图像经过方向滤波和低通滤波后各像素的自相关矩阵行列式值,以此构造特征角点图像;然后采用OTSU算法计算特征角点图像分割阈值,从而筛选出预选区域;最后结合改进的非极大值抑制方法提取有效角点.通过5组角点检测对比试验结果数据分析,不同类型图像的角点检测准确率均有提高,高分二号遥感影像的角点检测准确率提高27.06个百分点,可以初步得出,该算法相比传统Harris角点检测算法不但能够自动计算角点检测的最佳阈值,而且能够更准确地定位角点和去除边缘伪角点,从而提高了角点检测的精确度,该研究可为农业遥感影像数据检测提供参考.     

运用模糊隶属度进行土壤属性制图的研究——以黑龙江鹤山农场研究区为例

通过传统土壤类型图所得的土壤属性图已不能满足精准农业和生态环境模型所需土壤属性的精度,而目前应用较多的统计方法和地统计方法均存在一定的局限性。鉴于此,本文探索了一种采用模糊聚类获取模糊隶属度进行土壤属性制图的方法。首先,采用模糊c均值聚类(Fuzzyc-means clustering,FCM)方法对环境因子进行聚类,通过野外采样(称为建模点)建立土壤-环境关系知识;然后,计算区域内各像元点对土壤类型的模糊隶属度;最后,对模糊隶属度采用加权平均的方法获取土壤属性值。将该方法应用于黑龙江鹤山农场老莱河流域的研究小区,以土体厚度和表层有机质为例进行土壤属性制图。为了评价该方法的有效性,将其与采用环境因子所建立的多元线性回归模型进行比较,通过野外验证点集评价两种模型所得的土壤属性,评价指标为观测值和预测值的相关系数、平均绝对误差(MAE)、均方根误差(RMSE)和准确度(AC)。结果表明,尽管通过建模点建立的多元线性回归方程R2较大,但该方程并不适用于研究区内的其他样本点,这表明多元线性回归方法在该区具有一定的局限性。与之相比,模糊隶属度加权平均的方法则可以通过较少的建模点得到更好的预测效果。     

改进烟花算法和概率神经网络智能诊断齿轮箱故障

针对复杂环境下农机设备的齿轮箱系统在故障诊断时存在易受现场噪声干扰和故障识别率低等问题,提出了一种基于改进的烟花算法和概率神经网络的齿轮箱智能故障诊断方法。为提高现有概率神经网络模式分类方法的性能,定义了一项样本相似度衡量指标以提高建模过程中训练样本的质量。将烟花算法与概率神经网络技术有机融合提出了一种改进的烟花算法-概率神经网络模式分类方法,利用烟花算法优化概率神经网络的平滑参数以确定网络参数的最优值,提高模式分类与识别精度。将改进的烟花算法-概率神经网络模式分类方法用于噪声环境下齿轮箱的故障诊断建模,构建故障特征参量与齿轮箱工作状况间的复杂非线性映射关系。应用结果表明,与基于BP神经网络、GABP(genetic algorithm back propagation)神经网络和概率神经网络的故障诊断模型相比,在不同程度噪声影响下烟花算法-概率神经网络模型均具有最高故障识别率。当噪声控制系数为0.01、0.02、0.04和0.06时,模型的故障识别率分别为100%、95.83%、93.33%和88.33%。该研究可为非线性复杂系统的故障诊断提供了一种可行的解决方案。     

基于点特征检测的农业航空遥感图像配准算法

针对当前无人机遥感图像配准算法普遍存在匹配精度差与配准速度慢等问题,该文以点特征检测方法为基础,结合矩阵降维处理方法,提出一种适用于农业航空遥感图像配准的改进算法—SNS(scale-invariant feature transform and singular value decomposition)算法。SNS算法以高斯函数同步检测尺度空间极值点的坐标和特征尺度,利用海森矩阵消除伪特征点,获取特征点精准定位,在求取特征点的模值与方向基础上,采用奇异值分解方法进行矩阵优化,实现数据降维再重构。试验结果表明,SNS算法与经典算法相比,配准速度平均提高5.01%,配准精度均方根误差平均降低10.48%,说明SNS算法在压缩数据量的同时,提高了整体配准精度,具有配准速度较快和鲁棒性较好的特点。研究结果可为农业航空遥感图像快速配准提供参考。     

基于改进的粒子群多阈值算法的白色异性纤维检测

为了提高皮棉中白色异性纤维的识别精度,该文提出了一种基于改进混沌粒子群的白色异性纤维检测算法,该算法将图像的像素点按灰度值分为多类,把所有相邻类间方差看做一个粒子种群,以最大类间方差组作为种群适应度评价函数。通过滑动窗口技术判断算法是否陷入局部最优。有效克服了标准粒子群算法容易陷入局部最优的缺陷。通过试验验证,该文提出的算法对白色异性纤维的识别准确率达到98.6%。通过与标准二维Otsu算法的对比分割试验发现在分割较细小的白色异性纤维以及白色纤维与皮棉发生重叠的情况时,该算法的分割结果比标准二维Otsu算法更准确,噪声点更少。为皮棉异性纤维检测与剔除工艺的改善提供了技术依据。     

基于流形学习算法的马铃薯机械损伤机器视觉检测方法

针对马铃薯表面芽眼和凹凸不平的影响,使之马铃薯机械损伤难以检测的问题,该文提出了一种基于流形学习算法的马铃薯机械损伤检测方法。首先利用马铃薯图像的显著图分割出马铃薯区域,然后利用主成分分析(principal component analysis,PCA)、等距映射(isometric mapping,Isomap)和局部线性嵌入(locally-linear embedding,LLE)3种流形学习方法提取马铃薯区域图像特征参数,然后分别建立基于3种流形特征的支持向量机(support vector machine,SVM)分类模型PCA-SVM、Isomap-SVM和LLE-SVM,利用网格搜索法(grid search)、遗传算法(genetic algorithm,GA)以及粒子群算法(particle swarm optimization,PSO)3种模型参数优化方法,优化支持向量机模型的惩罚参数c和RBF核参数g,以建立最优分类模型,最后比较3种分类模型的识别效果,确定最优分类模型。研究结果表明,PCA-SVM分类模型对训练集识别率为100%,测试集识别率为100%;Isomap-SVM分类模型对训练集识别率为100%,测试集识别率为91.7%;LLE-SVM分类模型对训练集识别率为100%,测试集识别率为91.7%,表明PCA、Isomap和LLE 3种流形学习方法用于马铃薯机械损伤检测是可行的,其中PCA-SVM分类模型检测效果最优。     

A headspace solid-phase microextraction method for the determination of some secondary compounds of Brazilian sugar cane spirits by gas chromatography

A headspace solid-phase microextraction (SPME) method was developed for the determination of secondary compounds from Brazilian sugar cane spirits, or cacha?a, by GC-FID. An SPME holder with an 85 microm polyacrylate coating was utilized. The novel method is compared with an optimized method: liquid-liquid extraction (LLE). Both methods showed good linearity, but the repeatability for analyses done with the SPME technique (%RSD = 1.8-3.9) was better than for those done with LLE (%RSD = 10.3-11.7). The concentrations of the analytes obtained in the analysis of 12 cacha?a samples with the SPME technique were higher than those obtained with LLE. In the SPME method the extraction wastes are smaller. Cacha?a samples were qualitatively analyzed for GC-MS.     

刀具磨损声发射信号的混沌特性分析

金属切削是一个非线性系统,刀具磨损产生的声发射信号具有混沌特征。该文采用混沌理论对刀具不同磨损程度的声发射信号进行非线性特性分析。首先采用经验模态分解与小波阈值结合(empirical mode decomposition and wavelet,EMD-Wavelet)的法对信号去噪,消除噪声对吸引子结构以及特征参数的影响。其次利用互信息法和Cao方法分别求时延和嵌入维,对去噪后的信号进行相空间重构。最后绘制吸引子相图,并求解特征参数关联维。结果表明,看似无序杂乱的非线性声发射信号有着内在的有序状态。吸引子结构随着刀具磨损状态的改变呈现一定变化规律,关联维数与刀具磨损状态有一定的对应关系。这些特性对于刀具磨损状态识别有一定的参考意义。     

Threshold effect of eucalypt density on an aggressive avian competitor

General guidelines available to revegetation planners focus on the spatial context and dimensions of the revegetated site. However, site-specific habitat factors can have overarching importance for habitat value, especially where interactions with competitors or predators may play an important role. Current revegetation projects in Australia which aim to restore slow-growing buloke (Allocasuarina luehmannii) woodland, a threatened habitat important for bird conservation, usually include faster-growing eucalypts in plantings. This research aimed to identify whether eucalypt presence in buloke woodland facilitates invasion by the noisy miner (Manorina melanocephala), an aggressive competitor that is absent from pure buloke woodland. Birds were surveyed in buloke woodland remnants that contained eucalypts as a sub-dominant species at densities of 0-16 per ha. The probability of noisy miner presence in buloke woodland increased markedly where eucalypts were present at a density of approximately five per hectare. The presence of noisy miners resulted in a substantial difference in bird assemblage structure and composition. Small-bodied insectivorous birds which are experiencing population declines in southern Australia were recorded on average six times more often in transects without noisy miners (low-eucalypt density transects). Avian behaviour and habitat use was also altered, with birds flying more frequently in transects where noisy miners were present. A minor difference in habitat composition results in substantial degradation of the conservation value of non-eucalypt woodland in eastern Australia due to invasion by aggressive avian competitors. Revegetation and restoration practices should take into account the potential for such subtle floristic differences to result in substantial variation in conservation outcomes.     

Multiresidue method for determination of 35 pesticides in virgin olive oil by using liquid-liquid extraction techniques coupled with solid-phase extraction clean up and gas chromatography with nitrogen phosphorus detection and electron capture detection

A method for the multiresidue determination of 35 pesticides (30 insecticides and five herbicides) in olive oil by gas chromatography (GC) is described. Three liquid-liquid extraction (LLE) procedures based on (i) partition of pesticides between acetonitrile (ACN) and oil solution in n-hexane, (ii) partition of pesticides between saturated ACN with n-hexane and oil solution in n-hexane saturated with ACN, and (iii) partition of pesticides between ACN and oil were tested for the optimization of the highest pesticide recoveries with the lowest oil residue in the final extracts. Experimental tests were preformed in order to study the efficiency of different clean up procedures with N-Alumina, Florisil, C18, and ENVI-Carb solid-phase extraction (SPE) cartridges for the compounds analyzed by GC-nitrogen phosphorus detection. A second step of clean up was also performed for the compounds analyzed by GC-electron capture detection (ECD), by using phenyl-bonded silica (Ph), diol-bonded silica (Diol), cyanopropyl-bonded silica (CN), and amino propyl-bonded silica (NH2) SPE cartridges. LLE of the oil solution in hexane with ACN followed by an ENVI-Carb SPE clean up of the extract gave the best results for all target compounds. The ACN extract was additionally cleaned through a Diol-SPE cartridge for the determination of pesticides analyzed mainly by GC-ECD. Pesticide recoveries form virgin olive oil spiked with 20, 100, and 500 microg/kg concentrations of pesticides ranged from 70.9 to 107.4%. The proposed method featured good sensitivity, pesticide quantification limits were low enough, and the precision, expressed as relative standard deviation, ranged from 2.4 to 12.0%. The proposed method was applied successfully for the residue determination of the selected pesticides in commercial olive oil samples.     

Rapid determination of ractopamine in swine urine using surface-enhanced Raman spectroscopy

Ractopamine is approved for use in swine to improve carcass leanness in the United States, but banned in the European Union and China because ractopamine residue may pose health risks. This study investigated the possibility of applying surface-enhanced Raman spectroscopy (SERS) for analysis of ractopamine in swine urine. Ractopamine (0.1-10 μg mL(-1)) was added to urine samples collected from 20 swine to prepare a total of 240 samples. A simple centrifugation, a liquid-liquid extraction (LLE) method, and a more complicated method involving liquid-liquid extraction and solid-phase extraction (LLE-SPE) were used to extract ractopamine from urine samples. Principal component analysis (PCA) and partial least-squares (PLS) regression were used for spectral data analyses. Although no satisfactory result was obtained with the centrifugation method, ractopamine could be detected at levels of 0.8 and 0.4 μg mL(-1) with the LLE and LLE-SPE extraction methods, respectively. The R2 of the PLS model of actual ractopamine values versus predicted values was 0.74 for the LLE method and 0.73 for the LLE-SPE method. The SERS method with simple sample preparation has great potential for rapid analysis of ractopamine in swine urine.     

结合非子采样轮廓变换和形态收缩算子的多源遥感影像配准

针对多源遥感影像自动配准中难以提取大量同名特征点的问题,提出了一种结合非子采样轮廓变换和形态收缩算子的自动配准算法。结合非子采样轮廓变换和形态收缩算子的特征提取算法能够克服角度和尺度偏差,在多方向、多尺度空间精确提取强边缘上的关键结构特征点;基于低频波段的归一化互信息匹配算法和三角形一致检验算法能够提取到大量高可靠性的同名特征点对,保证了多源遥感影像的高精度配准。文中选取角度和尺度偏差显著的SPOT-5(P)和ASTER影像组合进行试验,结果证明以上算法能够检测到大量分布均匀的同名特征点对,配准模型精度趋近于1个像元。该研究可为多源遥感数据的融合和目标识别提供前提条件。     

基于运动恢复结构算法的油菜NDVI三维分布

植物形态伴随着植物生长过程而发生变化,植物的三维重建对研究植物形态对植物生物量估测、植物病害虫害、基因型表达等有着很重要的意义。目前三维重建方法重建出的三维点云多包含植物的形态、颜色等特征,无法反应植物营养状况（如叶绿素含量）、病虫害胁迫等原因造成有机质空间三维分布改变,同时以往手段都需要专门仪器,携带和作业都受到很大限制。多光谱图像能够反应有机质含量等化学值的分布,在近地面遥感、农产品质量无损检测等发面取得了广泛的应用。该文通过采集31张4叶龄油菜的多光谱图像,使用运动恢复结构算法（structure from motion）方法对其进行空间三维重建,得到油菜的三维点云,并对点云中噪声点进行滤除。以控制点和控制长度对所得模型进行评价,得到长度最大偏差在0.1023 cm,RMSE=0.052599,证明该方法重建所得模型具有较好的空间均匀性与准确性,最后计算NDVI指数空间分布。证明所得模型对将来研究植物营养与病虫害胁迫空间分布有着重要意义。     

Outliers in collaborative studies: coping with uncertainty

An overview is given of the options available in detecting and dealing with outliers in collaborative studies. The fundamental points of agreement and disagreement are highlighted. The common sense approach of just looking at the data is emphasized. The importance is stressed of making a harmonized choice of outlier treatments, even though such a choice may not be optimal for all circumstances.