高光谱图像技术在水果品质无损检测中的应用

传统的近红外光谱分析法和可见光图像技术应用于水果品质无损检测中存在的检测区域小、检测时间长、仅能检测表面情况等局限性。高光谱图像技术结合光谱技术与计算机图像技术两者的优点,可获得大量包含连续波长光谱信息的图像块,其图像信息可检测水果的外部品质,光谱信息则可用于水果内部品质的检测,达到根据水果内、外部综合品质进行分类的目的。根据不同的采集设备,简介了两种获得高光谱图像的方法。综述了国内外将该技术应用于水果品质检测方面的研究进展,检测内容包括外观品质、损伤与缺陷,成熟度和坚实度,含糖量、含水率等内部品质,着重介绍了各高光谱图像的成像波段范围、分辨率、成像源,实验数据处理的方法以及实验结果等。根据综述所得提出了高光谱图像技术应用中需要解决的光谱降维、降低样品差异影响和实时检测平台搭建等问题。     

基于高光谱的冬油菜植株氮素积累量监测模型

为无损和定量研究高光谱技术在冬油菜植株氮素积累量(PNA,plant nitrogen accumulation)时空变化监测的适宜性及准确性,该文以两年田间氮肥水平试验为基础,采用单变量线性和非线性回归方法,建立基于特征光谱参数的冬油菜P NA高光谱估算模型。结果表明,采用比值光谱的方法可显著提高冬油菜冠层光谱反射率与PNA间的相关性,其最佳的波段组合为1 259 nm与492 nm处光谱反射率比值(R1259/R492),决定系数R2为0.85。高光谱参数间,以比值植被指数(RVI-5)、归一化光谱指数(NDSI)、线性内插法红边位置(REIP)、三角植被指数(TVI)、742 nm处一阶微分光谱值(FD742)和红边面积(SDR)等光谱参数与PNA相关性较好(平均R2和标准误SE分别为0.69和42.70),且以FD742表现最优(R2=0.79,SE=35.66)。精度分析结果显示,以光谱参数R1259/R492和FD742为自变量的指数方程模型作为高光谱监测油菜PNA的最佳模型,各生育期Noise Equivalent(NE)均较低且表现稳定,同时模型估测精度较高,R2分别为0.98和0.98,相对均方根误差RRMSE分别为0.73和0.72,相对误差MRE分别为14.42%和10.31%。该方法为快捷和精确评估冬油菜PNA提供了新的研究思路。     

玉米苗期冠层多光谱反射率反演与叶绿素含量诊断

为了探索玉米苗期叶片叶绿素含量指标的快速、非破坏性估测方法,该文运用多光谱图像技术对大田玉米苗期叶绿素含量指标进行快速无损的诊断研究。大田试验中,采用2-CCD多光谱图像采集系统获取大田玉米苗期的冠层多光谱图像,并同步采集漫反射灰度板的多光谱图像。为消除光照对图像采集质量的影响,准确将不同光照条件下的玉米冠层图像数据转换为其叶面反射率数据,标定试验中采用一块4个不同灰度级的满足朗伯面条件漫反射灰度板,建立了叶片光谱反射率同图像灰度值之间的线性反演公式,并与大田试验中漫反射灰度板的多光谱图像建立了玉米冠层图像灰度值的校正公式。对玉米苗期冠层多光谱图像进行处理,提取出玉米冠层B、G、R、NIR(中心波长分别为470,550,620,800 nm)4个波段归一化平均灰度值。通过灰度值的校正公式得到校正后的归一化平均灰度值,由线性公式反演出R、G、B、NIR 4个波段的平均反射率值,并计算4种常见光谱植被指数(RNDVI、RNDGI、RRVI和RDVI),采用最小二乘-支持向量回归(LS-SVR)建立植被指数同叶绿素含量指标的拟合模型。结果表明:植被指数RNDVI、RRVI和RDVI和玉米冠层叶绿素含量指标拟合验证集决定系数R2为0.56,达到了较为理想的拟合结果。证明通过漫反射灰度板对玉米冠层多光谱图像建立反射率反演校正模型的方法是可行的,这一方法为快速无损检测玉米苗期叶绿素含量指标提供了支持。     

COMPARISON OF FLUID-ATTENUATED INVERSION RECOVERY AND T2-WEIGHTED MAGNETIC RESONANCE IMAGES IN DOGS AND CATS WITH SUSPECTED BRAIN DISEASE

To compare fluid-attenuated inversion recovery (FLAIR) and T2-weighted magnetic resonance (MR) imaging in small animal patients with suspected brain disease, paired sets of FLAIR and T2-weighted MR images of 116 dogs and cats were reviewed separately without any patient information. Images were rated as normal or abnormal using a five-point scale, and the distribution, signal intensity, and anatomic location of abnormalities were recorded. In 60 animals, both FLAIR and T2-weighted images were normal. In 50 animals, the same abnormalities were identified in both FLAIR and T2-weighted images. Overall, very good agreement was found between FLAIR and T2-weighted MR images (kappa = 0.88). FLAIR images had abnormalities that were not recognized in the corresponding T2-weighted images in six of 116 examinations (5%). In four of these, the abnormalities in FLAIR images were thought to represent pathology, including granulomatous meningoencephalitis in one dog, postictal edema in one dog, and undiagnosed lesions in two dogs. In the remaining two examinations, the abnormalities in FLAIR images were probably artifacts. No examples were found of intracranial abnormalities in T2-weighted images that were not visible in FLAIR images. In this study, acquiring FLAIR images in addition to T2-weighted images resulted in detection of otherwise occult abnormalities in relatively few patients.     

Development of a spot plow providing complete inversion for effective weed control

Tillage for the "complete inversion" of soil, that is, overturning soil slices 180° was proposed, a "spot plow" was developed and tested to accomplish the task, and a simulation model was evaluated to demonstrate the efficacy of the plow on weed control. A 360 mm wide spot plow was designed to operate at a speed of 1.9 m s⁻¹ for the spot plowing with the least possible lateral displacement of the soil slice by utilizing the inertia of the soil slice and securely rotating it. In field experiments, complete spot inversion required an operating speed of at least 1.6 m s⁻¹; at lower speeds, a portion of the soil block was left half-inverted and further lowering led to considerable lateral displacement. The displacement in the forward and lateral directions was minimal, implying that spot plowing is suitable for potential application to and verification of the weed population dynamics model in the field. A simple linear matrix model of the population dynamics of annual weeds was proposed, whereby four layers of soil were set to describe tillage and other ecological events. The effect of tillage on weed control was evaluated by the equilibrium reproduction rate allowed to sustain a stable population of weeds. The simulation model showed that alternately changing the depth of spot plowing had a significant effect on controlling weeds of low-survival-rate seeds, even when some incomplete inversion of the soil slice was taken into account.     

冬小麦生物量及氮积累量的植被指数动态模型研究

利用遥感技术实时监测小麦生长状况,依据监测结果适时促控,可提高产量。本研究以高产小麦品种周麦27为试验材料,在不同试验地点设置了水氮耦合的大田试验,筛选出了适宜监测冬小麦地上部氮积累量和生物量的植被指数,并构建了不同产量水平下优选植被指数的动态模型。结果表明,(1)不同的水氮耦合模式显著影响小麦冠层光谱变化,在350~700nm和750~900nm表现相反的反应特征;(2)对2个农学生长指标反应敏感且兼容性好的植被指数主要有修正型红边比率(mRER)、土壤调整植被指数[SAVI(825,735)]、红边叶绿素指数(CIred-edge)和归一化差异光谱指数(NDSI),其与产量间相关性较好的时期为拔节至灌浆中期;(3)双Logistic模型可以很好地拟合植被指数的动态变化,高产和超高产水平下拟合精度较高(R^2>0.82),而低产水平下相对较低(R2=0.608~0.736)。比较而言,CIred-edge和SAVI(825,735)用于评价小麦长势较为适宜。研究结果对作物因地定产、以苗管理、分类促控具有重要意义。     

水稻冠层二向反射率的模拟及其反演

 在2000年晚稻大田试验中，设5个氮素水平，即严重缺氮、缺氮、氮适中、氮过多和氮严重过多5种不同的生长状况，实测了5种氮素水平下的水稻冠层二向反射率、叶片反射率、透射率、冠层结构参数等，利用FCR模型模拟冠层二向反射率，模拟精度用RMSE值和相关系数r检验，利用POWELL法反演冠层结构参数，当叶面积指数>3时，反演结果与实测值的相对误差小于10%。     

基于RGB图像光谱重建的鱼糜掺假定量检测研究

实现基于RGB图像的光谱重建对降低光谱的硬件要求、扩大其实际应用具有重大意义。该研究以鱼糜掺假检测为例,比较多元多项式最小二乘回归算法（polynomial multivariate least-squares regression,PMLR）与深度学习HRNet网络对光谱重建的性能,建立基于重建光谱多种掺假鱼糜检测模型并验证其实际应用的有效性。结果表明,2种方法的重建光谱误差较小,HRNet网络、PMLR算法重建光谱的均方根误差（root mean square error,RMSE）分别为0.010 4和0.012 6,大多数掺假检测模型有较高的预测准确性,其预测相关系数大于0.91,预测均方根误差小于9%。在基于重建光谱建立的掺假检测模型中,效果最佳的是基于PMLR算法重建光谱使用标准正态变量变换（standard normal variate,SNV）预处理的极限学习机回归模型,其预测均方根误差为3.954 4%、预测相关系数为0.983 0。因此,PMLR算法和HRNet网络均能较好的实现基于RGB图像的光谱重建,且重建光谱均能实现对鱼糜掺假样本的较好检测结果,为基于重建光谱的食品和农产品品质与安全检测提供了新思路。     

基于野外VIS-NIR光谱的土壤盐分主要离子预测

为明确干旱区土壤盐分主要离子的特征光谱,建立精度高和稳定性好的盐渍土预测模型,以新疆阜康市为研究区域,采用网格法采集55个土壤样本,利用实测VIS-NIR光谱,选择多元线性回归(MLR)、支持向量机(SVM)、随机森林(RF)法构建土壤盐分主要离子含量反演模型,而后对反演精度进行检验。结果显示:(1)在0.01显著水平下,土壤盐分与Na⁺、Cl^–、Ca²⁺含量均呈显著相关,相关系数分别为0.978、0.814、0.645;(2)综合光谱响应和相关性分析确定土壤盐分主要离子的特征波段为459、537、1 381、1 386 nm,显著特征波段为459、537 nm;(3)3种模型拟合效果从高到低依次为RF>MLR>SVM,采用RF所建模型盐分主要离子(Na⁺、Cl^–、Ca²⁺)R²最高,RMSE最小,RPD最大,分别为2.11、2.03、1.80,为最优预测模型。通过选取土壤主要离子显著特征波段,进而采用RF法构建其估测模型...     

油菜红边特征及其叶面积指数的高光谱估算模型

在2003～2004年油菜生长季,选用6个油菜品种,设置3个氮素水平的小区试验。在不同发育期同步测定油菜冠层的光谱反射率及对应叶片的叶面积指数。利用油菜冠层的光谱反射率数据提取红边参数,分析其变化规律,油菜叶面积指数与红边参数的相关性,估算结果表明：油菜冠层红边一阶导数光谱具有“双峰” 现象,红边位置λ_red位于690～720 nm之间,在油菜生长旺盛期间出现“红边平台”,前期有“红移”,后期有“蓝移”现象;叶面积指数与冠层光谱红边参数λ_red、Dλ_red、S_red之间在开花前存在显著相关,但开花后相关性不显著;利用开花前的红边参数可以估算油菜的叶面积指数,开花后的红边参数不能用于估算油菜的叶面积指数;最后建立了不同时期和开花前油菜叶面积指数的估算模型。