Detection of Thrips Defect on Green-Peel Citrus Using Hyperspectral Imaging Technology Combining PCA and B-Spline Lighting Correction Method

In order to find an effective method of detecting thrips defect on green-peel citrus, a defect segmentation method was developed using a single threshold value based on combination of characteristic wavelengths principal component analysis （PCA） and B-spline lighting correction method in this study. At first, four characteristic wavelengths （523, 587, 700 and 768 nm） were obtained using PCA of Vis-NIR （visible and near-infrared） bands and analysis of weighting coefficients; secondarily, PCA was performed using characteristic wavelengths and the second principal component （PC2） was selected to classify images; then, B-spline lighting correction method was proposed to overcome the influence of lighting non-uniform on citrus when thrips defect was segmented; finally, thrips defect on citrus was extracted by global threshold segmentation and morphological image processing. The experimental results show that thrips defect in citrus can be detected with an accuracy of 96.5% by characteristic wavelengths PCA and B-spline lighting correction method. This study shows that thrips defect on green-peel citrus can be effectively identified using hyperspectral imaging technology.     

基于高光谱成像技术的红酸枝木材种类识别

为了实现市场上常见红酸枝类Dalbergia spp.木材的快速无损识别,利用高光谱成像技术对不同红酸枝木材进行种类识别研究。以交趾黄檀 Dalbergia cochinchinensis,巴里黄檀 Dalbergia bariensis,奥氏黄檀Dalbergia oliveri和微凹黄檀 Dalbergia retusa为研究对象,采集高光谱图像并提取感兴趣区域内的反射光谱,采用Savitsky-Golay（SG）平滑算法、标准正态变量变换（SNV）和多元散射校正（MSC）对955~1 642 nm 波段光谱进行预处理,并通过主成分分析法（PCA）,回归系数法（RC）以及连续投影法（SPA）选择特征波长,分别建立了偏最小二乘判别分析（PLS-DA）和极限学习机（ELM）判别分析模型。研究结果表明:经SG和MSC光谱预处理,采用SPA选择的特征波长建立的ELM模型性能最优,建模集和预测集的识别率均为100.0%。这为红酸枝木材种类的快速无损识别提供了新的方法。图5表4参17     

锦橙叶片钾含量光谱监测模型研究

【目的】快速、无损、准确地获取柑橘叶片营养信息。【方法】以盆栽蓬安100锦橙为试材,通过精确控制施肥处理(K0:0g,K1:30g,K2:75g,K3:90,K4:120gk2O/株/年),利用鲜叶进行光谱检测钾素营养状况分析。【结果】可见近红外波段范围内,各施钾处理蓬安100锦橙夏梢叶片光谱反射强度呈K3K0K1K2K4趋势。通过对反射光谱、一阶微分、二阶微分和倒数对数光谱进行多元散射(multiple scattering correction,MSC)校正处理,运用偏最小二乘法(partial least square method,PLS)与内部交叉验证建立了钾含量预测回归模型,其中反射光谱的二阶微分光谱钾含量定标模型具有最好的预测能力,其预测相关系数最大,r=0.82;预测均方根误差较小,RMSEP=0.0038;偏差(Bias)绝对值最小,Bias=-2.34E-05。【结论】通过锦橙叶片反射光谱二阶微分值与叶片钾含量构建的PLS回归模型,可以较好地预测蓬安100锦橙夏梢叶片钾含量。进一步分析表明,波段477—515nm、541—588nm、632—669nm、701—718nm和754—794nm是反射光谱二阶微分与蓬安100锦橙叶片钾含量定标模型的特征波长。     

Detection of Fusarium damaged kernels in Canada Western Red Spring wheat using visible/near-infrared hyperspectral imaging and principal component analysis

Fusarium damage in wheat reduces the quality and safety of food and feed products. In this study, the use of hyperspectral imaging was investigated to detect fusarium damaged kernels (FDK) in Canadian wheat samples. Eight hundred kernels of Canada Western Red Spring wheat were segregated into three classes of kernels: sound, mildly damaged and severely damaged. Singulated kernels were scanned with a hyperspectral imaging system in the visible-NIR (400-1000 nm) wavelength range. Principal component analysis (PCA) was performed on the images and the distribution of PCA scores within individual kernels measured to develop linear discriminant analysis (LDA) models for predicting the extent of fusarium damage. An LDA model classified the wheat kernels into sound and FDK categories with an overall accuracy of 92% or better. Classification based on six selected wavelengths was comparable to that based on the full-spectrum data.     

基于光谱与图像信息的杏鲍菇多糖含量检测

【目的】利用高光谱成像技术实现杏鲍菇Pleurotus eryngii多糖含量的快速无损检测。【方法】利用高光谱图像采集系统获取350～1 021 nm波长范围内的杏鲍菇高光谱图像,同时利用苯酚–硫酸法测定对应样本的多糖含量。通过波段运算和阈值分割构建掩膜图像,使样本与背景相分离。采用主成分分析(PCA)处理原始高光谱图像,获得代表原始图像99%信息的2个主成分图像(PC1、PC2),然后利用连续投影算法(SPA)选出554.4、772.8、811.4、819.1、855.6、986.3和1 019.5 nm 7个特征波长及对应的光谱特征,分别提取7个特征波长图像和2个主成分图像的纹理与颜色特征,最后利用偏最小二乘回归(PLSR)建立杏鲍菇样本基于不同图像特征与多糖含量之间的关系模型。【结果】从校正集决定系数(Rc2)来看,基于特征光谱+特征波长图像特征+主成分图像特征的模型效果最好,Rc2=0.954,RMSEc=0.341;从预测集决定系数Rp2来看,基于特征光谱+特征波长图像特征的模型效果最好,Rp2=0.868,RMSEP=0.539。【结论】该研究结果可为杏鲍菇多糖含量的快速、无损检测提供一定的参考。     

Application of neural networks to discriminate fungal infection levels in rice panicles using hyperspectral reflectance and principal components analysis

Detecting plant health condition is an important step in controlling disease and insect stress in agricultural crops. In this study, we applied neural network and principal components analysis techniques for discriminating and classifying different fungal infection levels in rice (Oryza sativa L.) panicles. Four infection levels in rice panicles were used in the study: no infection condition, light and moderate infection caused by rice glume blight disease, and serious infection caused by rice false smut disease. Hyperspectral reflectance of rice panicles was measured through the wavelength range from 350 to 2500 nm with a portable spectroradiometer in the laboratory. The spectral response characteristics of rice panicles were analyzed, and principal component analysis (PCA) was performed to obtain the principal components (PCs) derived from different spectra processing methods, namely raw, inverse logarithmic, first, and second derivative reflectance. A learning vector quantization (LVQ) neural network classifier was employed to classify healthy, light, moderate, and serious infection levels. Classification accuracy was evaluated using overall accuracy and Kappa coefficient. The overall accuracies of LVQ with PCA derived from the raw, inverse logarithmic, first, and second derivative reflectance spectra for the validation dataset were 91.6%, 86.4%, 95.5%, and 100% respectively, and the corresponding Kappa coefficients were 0.887, 0.818, 0.939 and 1. Our results indicated that it is possible to discriminate different fungal infection levels of rice panicles under laboratory conditions using hyperspectral remote sensing data.     

基于高光谱成像技术的茄子叶片灰霉病早期检测

为建立基于高光谱成像技术的茄子叶片灰霉病早期检测方法,利用高光谱成像系统获取120个茄子叶片在380~1031nm范围的高光谱图像数据,通过主成分分析(PCA)对高光谱数据进行降维,并从中优选出3个特征波段下的特征图像,截取200×150的感兴趣区域图像(ROI),并从每幅特征图像中分别提取均值、方差、同质性、对比度、差异性、熵、二阶矩和相关性等8个基于灰度共生矩阵的纹理特征变量,通过连续投影算法(SPA)提取13个特征变量, 利用最小二乘支持向量机(LS‐SVM)构建茄子叶片灰霉病早期鉴别模型,模型判别准确率为97.5%.说明高光谱成像技术可以用于茄子叶片灰霉病的早期检测.     

基于主成分分析和BP神经网络的土壤养分近红外光谱检测

基于近红外光谱技术的土壤养分快速、无损检测,有利于精细施肥决策。在一黄豆田采用7 m×7 m的栅格采集54个土样,测定其土壤有机质、速效氮、有效磷、有效钾,并使用FieldSpec 3光谱仪测定土样的近红外漫反射光谱。将54个样本随机分成预测集与验证集,其中预测集40个,验证集14个。通过平滑预处理后,利用主成分分析法(PCA)提取原始光谱8个主成分。然后以8个主成分为输入,分别以所测土壤养分作为输出,建立土壤有机质、速效氮、有效磷、有效钾的预测模型,最后对14个验证样本进行预测。结果表明,在小尺度采样的情况下进行光谱分析,采用主成分分析和人工神经网络相结合的方法建立土壤有机质预测模型,其测量值与预测值的相关性较高,相关度为0.796 2,相对误差较小,其平均值为1.88%,表明该方法预测土壤有机质含量是可行的。但对土壤速效氮、有效磷和有效钾含量的预测并不理想,还有待进一步研究。     

Using hyperspectral imaging to discriminate yellow leaf curl disease in tomato leaves

This paper investigated the possibility of discriminating tomato yellow leaf curl disease by a hyperspectral imaging technique. A hyperspecral imaging system collected hyperspectral images of both healthy and infected tomato leaves. The reflectance spectra, first derivative reflectance spectra and absolute reflectance difference spectra in the wavelength range of 500–1000 nm of both background and the leaf area were analyzed to select sensitive wavelengths and band ratios. 853 nm was selected to create a mask image for background segmentation, while 720 nm from the reflectance spectra, four peaks (560, 575, 712, and 729 nm) from the first derivative spectra and, four wavelengths with higher values (586, 720 nm) and lower values (690, 840 nm) in the absolute difference spectra were selected as a set of sensitive wavelengths. Four band ratio images (560/575, 712/729, 586/690, and 720/840 nm) were compared with four widely used vegetation indices (VIs). 24 texture features were extracted using grey level co-occurrence matrix (GLCM), respectively. The performance of each feature was evaluated by receiver operator characteristic (ROC) curve analysis. The best threshold values of each feature were calculated by Yonden’s index. Mean value of correlation (COR_MEAN) extracted from the band ratio image (720/840 nm) had the best performance, whose AUC value was 1.0. The discrimination result for a validation set based on its best threshold value was 100%. This research also demonstrated that multispectral images at 560, 575 and 720 nm have a potential for detecting tomato yellow leaf curl virus infection in field applications.     

基于高光谱的水稻叶片氮素营养诊断研究

为快速、准确地实现水稻氮素营养诊断,以中嘉早17水稻为试验对象,设置4种施氮水平的水稻栽培试验,利用便携式地物波谱仪获取240组水稻分蘖期顶三叶在350~2 500 nm的光谱数据。随机将样本划分为训练集(160个样本)和测试集(80个样本)。首先,通过多元散射校正(MSC)、变量标准化校正(SNV)、平滑算法(SG)3种方法分别对原始光谱进行预处理;然后,采用主成分分析(PCA)和连续投影算法(SPA)对预处理后的光谱进行特征降维,选取累积贡献率超过99.98%的前24个主成分作为模型的输入变量,对于经过MSC、SNV和SG处理后的光谱数据,还分别筛选出12、15、19个特征波长;最后,应用支持向量机(SVM)基于上述处理分别建立水稻氮素营养诊断模型。结果表明,采用MSC-PCA-SVM模型进行水稻氮素营养诊断的识别准确率最高,其在训练集和预测集上的准确率分别达99.38%和97.50%。     

基于高光谱图像技术的油菜籽品种鉴别方法研究

提出了一种采用高光谱图像技术结合人工神经网络对油菜籽品种进行鉴别的方法.采集多个品种油菜籽400～1 000 nm范围的高光谱图像数据,通过主成分分析法(PCA)获得主成分图像,确定特征波长;采用基于灰度直方图和灰度共生矩阵联合的统计方法从特征图像中提取纹理特征参数,应用人工神经网络建立油菜籽品种鉴别模型.结果表明,模型训练时品种判别率为93.75%,预测的判别率为91.67%.说明高光谱图像技术对油菜籽品种具有较好的分类和鉴别作用.     

Visible-near infrared spectroscopy for detection of Huanglongbing in citrus orchards

This paper evaluates the feasibility of applying visible-near infrared spectroscopy for in-field detection of Huanglongbing (HLB) in citrus orchards. Spectral reflectance data from the wavelength range of 350-2500 nm with 989 spectral features were collected from 100 healthy and 93 HLB-infected citrus trees using a visible-near infrared spectroradiometer. During data preprocessing, the spectral data were normalized and averaged every 25 nm to reduce the spectral features from 989 to 86. Three datasets were generated from the preprocessed raw data: first derivatives, second derivatives, and a combined dataset (generated by integrating preprocessed raw data, first derivatives and second derivatives). The preprocessed datasets were analyzed using principal component analysis (PCA) to further reduce the number of features used as inputs in the classification algorithm. The dataset consisting of principal components were randomized and separated into training and testing datasets such that 75% of the dataset was used for training; while 25% of the dataset was used for testing the classification algorithms. The number of samples in the training and testing datasets was 145 and 48, respectively. The classification algorithms tested were: linear discriminant analysis, quadratic discriminant analysis (QDA), k-nearest neighbor, and soft independent modeling of classification analogies (SIMCA). The reported classification accuracies of the algorithms are an average of three runs. When the second derivatives dataset were analyzed, the QDA-based classification algorithm yielded the highest overall average classification accuracies of about 95%, with HLB-class classification accuracies of about 98%. In the combined dataset, SIMCA-based algorithms resulted in high overall classification accuracies of about 92% with low false negatives (less than 3%).     

Use of ground based hyperspectral remote sensing for detection of stress in cotton caused by leafhopper (Hemiptera: Cicadellidae)

Detection of crop stress is one of the major applications of hyperspectral remote sensing in agriculture. Many studies have demonstrated the capability of remote sensing techniques for detection of nutrient stress on cotton with only few on pest damage but none so far on leafhopper (LH) severity. Subsequent to introduction of Bt cotton, leafhopper is emerging as a key pest in several countries. In view of its wide host range, geographical distribution and damage potential, a study was initiated to characterise leafhopper stress on cotton, identify sensitive bands, and derive hyperspectral vegetation indices specific to this pest. Cotton plants with varying levels of LH severity were selected from three locations across major cotton growing regions of India. About 57-58 cotton plants from each location exhibiting different levels of LH damage symptoms were selected. Reflectance measurements in the spectral range of 350-2500 nm were made using hyperspectral radiometer. Simultaneously chlorophyll (Chl) and relative water content (RWC) were also estimated from the selected plants. Reflectance from healthy and leafhopper infested plants showed a significant difference in VIS and NIR regions. Decrease in Chl a pigment was more significant than Chl b in the infested plants and the ratio of Chl a/b showed a decreasing trend with increase in LH severity. Regression analysis revealed a significant linear relation between LH severity and Chl (R² = 0.505^∗∗), and a similar fit was also observed for RWC (R² = 0.402^∗∗). Plotting linear intensity curves between reflectance at each waveband with infestation grade resulted in six sensitive bands that exhibited maximum correlation at different regions of the electromagnetic spectrum (376, 496, 691, 761, 1124 and 1457 nm). Regression analysis of several ratio indices formulated with two or more of these sensitive bands led to the identification of new leaf hopper indices (LHI) with a potential to detect leafhopper severity. These new indices along with 20 other stress related hyperspectral indices compiled from literature were further tested for their ability to detect LH severity. Two novel indices LHI 2 and LHI 4 proposed in this study showed significantly high coefficients of determination across locations (R² range 0.521 to 0.825^∗∗) and hence have the potential use for detection of leafhopper severity in cotton.     

基于高光谱成像技术识别苹果轻微损伤的有效波段研究

为了筛选出适用于开发苹果轻微损伤自动分级仪器的有效波段,以200个烟台富士苹果为对象进行研究。首先获取400～1 000 nm波长范围内完好和轻微损伤后0、0.5、1 h的苹果高光谱图像,然后提取完好与损伤样本感兴趣区域的平均光谱反射率数据,再利用载荷系数法（x LW）、连续投影法（SPA）和二阶导数（second derivative）法提取特征波长,分别提取3、9和20个特征波长,并根据特征波长建立基于遗传算法优化的BP神经网络（GA BP）和支持向量机（SVM）损伤识别模型。结果显示,三种基于特征波长提取方法建立的SVM模型对测试集的识别率（分别为77.50%、91.88%、96.88%）均高于BP GA模型（分别为75.63%、90.63%、93.75%）,因此,SVM被确定为最佳苹果轻微损伤识别模型。最后,利用每一特征波长分别作为变量建立SVM模型。结果发现,波段811 nm识别率达到90.63%,优于其他波段,被确定为苹果轻微损伤识别的最优波段。     

Hyper-spectral characteristics and classification of farmland soil in northeast of China

The physical and chemical heterogeneities of soils make the soil spectral different and complicated, and it is valuable to increase the accuracy of prediction models for soil organic matter(SOM) based on pre-classification. This experiment was conducted under a controllable environment, and different soil samples from northeast of China were measured using ASD2500 hyperspectral instrument. The results showed that there are different reflectances in different soil types. There are statistically significant correlation between SOM and reflectence at 0.05 and 0.01 levels in 550–850 nm, and all soil types get significant at 0.01 level in 650–750 nm. The results indicated that soil types of the northeast can be divided into three categories: The first category shows relatively flat and low reflectance in the entire band; the second shows that the spectral reflectance curve raises fastest in 460–610 nm band, the sharp increase in the slope, but uneven slope changes; the third category slowly uplifts in the visible band, and its slope in the visible band is obviously higher than the first category. Except for the classification by curve shapes of reflectance, principal component analysis is one more effective method to classify soil types. The first principal component includes 62.13–97.19% of spectral information and it mainly relates to the information in 560–600, 630–690 and 690–760 nm. The second mainly represents spectral information in 1 640–1 740, 2 050–2 120 and 2 200–2 300 nm. The samples with high OM are often in the left, and the others with low OM are in the right of the scatter plot(the first principal component is the horizontal axis and the second is the longitudinal axis). Soil types in northeast of China can be classified effectively by those two principles; it is also a valuable reference to other soil in other areas.     

Quantitative analysis of minerals and electric conductivity of red grape homogenates by near infrared reflectance spectroscopy

The use of near infrared (NIR) reflectance spectroscopy to measure the concentration of minerals and electric conductivity (EC) in red grape homogenates was investigated. Wine grape samples (n = 209) from two vintages, representing a wide range of varieties and regions were analysed by Inductively Coupled Plasma Optical Emission Spectrometry (ICPOES) for the concentrations of calcium (Ca), potassium (K), magnesium (Mg), phosphorus (P), sulphur (S), iron (Fe), and manganese (Mn) and scanned in reflectance in a NIR instrument (400-2500 nm). The spectra were pre-processed using multiple scatter correction (MSC) before developing the calibration models using partial least squares (PLS) regression and cross validation. Coefficients of determination in cross validation (R²) and the standard errors of cross validation (SECV) obtained were for Fe (0.60 and 1.49 mg kg⁻¹), Mn (0.71 and 0.41 mg kg⁻¹), Ca (0.75 and 60.89 mg kg⁻¹), Mg (0.84 and 12.93 mg kg⁻¹), K (0.78 and 285.34 mg kg⁻¹), P (0.70 and 40.19 mg kg⁻¹), S (0.88 and 14.45 mg kg⁻¹) and EC (0.87 and 7.66 mS). The results showed that Mg, S and EC in grape berries might be measured by NIR reflectance spectroscopy.     

Use of hyperspectral data to assess the effects of different nitrogen applications on a potato crop

This study was conducted to explore whether hyperspectral data could be used to discriminate between the effects of different rates of nitrogen application to a potato crop. The field experiment was carried out in the Central Potato Research Station, Jalandhar, on seven plots with different nitrogen (N) treatments. Spectral reflectance was measured using a 512-channel spectroradiometer with a range of 395–1075 nm on two different dates during crop growth. An optimum number of bands were selected from this range based on band–band r ², principal component analysis and discriminant analysis. The four bands that could discriminate between the rates of N applied were 560, 650, 730, and 760 nm. An ANOVA analysis of several narrow-band indices calculated from the reflectance values showed the indices that were able to differentiate best between the different rates of N application. These were reflectance ratio at the red edge (R_740/720) and the structure insensitive pigment index (SIPI). To estimate leaf N, reflectance ratios were determined for each band combination and were evaluated for their correlation with the leaf N content. A regression model for N estimation was obtained using the reflectance ratio indices at 750 and 710 nm wavelengths (F-ratio = 32 and r ² = 0.551, P < 0.000).     

基于近红外光谱与 KPCA-SVM鉴别淡水鱼肉

为实现淡水鱼品种的快速鉴别,采用近红外光谱分析技术建立7种淡水鱼鲜肉的快速鉴别模型。试验采集了鲢、草鱼、乌鳢、鲫、鲤、青鱼、鳙7种淡水鱼共772个鲜鱼肉样品的近红外光谱数据,分别考察标准正态变换(standard normalized variate,SNV)、多元散射校正(multiplicative signal correction,MSC)的预处理方法及核主成分分析(kernel principal component analysis,KPCA)和主成分分析(principal component analysis,PCA)的特征提取方法对支持向量机(support vector machine,SVM)判别模型的影响。结果显示,经SNV预处理和KPCA提取特征变量后,对未知样品的整体正确判别率达到92.68%。因此,采用近红外光谱技术结合化学计量学方法所建SVM模型可以实现淡水鱼品种的快速鉴别。     

应用可见/近红外光谱技术快速鉴别山西陈醋品种

为了实现对山西老陈醋品种的快速鉴别,应用可见/近红外光谱透射技术,结合化学计量学方法,进行了山西老陈醋品种的判别分类试验研究。对4个不同品种共240个山西老陈醋样品采集其光谱数据,结合主成分分析和神经网络技术分别对山西陈醋原始光谱、一阶微分光谱、二阶微分光谱进行了判别分析。结果表明:可见/近红外原始光谱结合主成分分析神经网络判别分析法的分析结果最优,校正集正确分类的百分比达92.1%,预测集达85.0%;二阶微分光谱分析结果最差。     

The vegetation classification in coal mine overburden dump using canopy spectral reflectance

The canopy spectral characteristics of typical plants in the overburden of the Fuxin coal mine dump were measured and analyzed. The reflectance of Leymus chinensis was affected by the soil, with a slight shift from green (550 nm) to the near infrared (NIR) region. Changes in chlorophyll and water absorption were not significant in the red (670 nm) and NIR bands, respectively. The reflectance curve trend for Artemisia lavandulaefolia was similar to those of Sophora japonica and Ulmus pumila, while the reflectance of S. japonica and U. pumila fluctuated in the NIR region (760-1200 nm), especially with greater water absorption around 930 and 1120 nm. In contrast, the reflectance of A. lavandulaefolia fluctuated slightly around 930 nm and a significant peak appeared at 1127 nm. In addition, the spectral reflectance of S. japonica was lower than for the other species in the visible band (400-700 nm). However, it was higher than for L. chinensis in the NIR region (780-1200 nm). Three classifiers, the self-organizing map (SOM), learning-vector quantization (LVQ), and a probabilistic neural network (PNN), were used to classify the vegetation and the results of all classifiers were compared based on total spectral reflectance data from 400 to 1200 nm. The PNN was the best classifier in terms of training and testing accuracy. The first difference reflectance was calculated, and the red edge parameter was able to classify the herbs (L. chinensis and A. lavandulaefolia) and the arbores (S. japonica and U. pumila) with an accuracy of 77 and 84%, respectively, although it did not perform as well for detail species. A mixing parameter matrix was built based on the sensitive wavelengths (550, 674, 810, 935, and 1125 nm), the vegetation indices (SAVI and NDGI), and the water absorption slope. High classification accuracy was obtained by applying the mixing parameter matrix. This method could be used for revegetation monitoring and in decision making.