Early detection and classification of plant diseases with Support Vector Machines based on hyperspectral reflectance

Automatic methods for an early detection of plant diseases are vital for precision crop protection. The main contribution of this paper is a procedure for the early detection and differentiation of sugar beet diseases based on Support Vector Machines and spectral vegetation indices. The aim was (I) to discriminate diseased from non-diseased sugar beet leaves, (II) to differentiate between the diseases Cercospora leaf spot, leaf rust and powdery mildew, and (III) to identify diseases even before specific symptoms became visible. Hyperspectral data were recorded from healthy leaves and leaves inoculated with the pathogens Cercospora beticola, Uromyces betae or Erysiphe betae causing Cercospora leaf spot, sugar beet rust and powdery mildew, respectively for a period of 21 days after inoculation. Nine spectral vegetation indices, related to physiological parameters were used as features for an automatic classification. Early differentiation between healthy and inoculated plants as well as among specific diseases can be achieved by a Support Vector Machine with a radial basis function as kernel.The discrimination between healthy sugar beet leaves and diseased leaves resulted in classification accuracies up to 97%. The multiple classification between healthy leaves and leaves with symptoms of the three diseases still achieved an accuracy higher than 86%. Furthermore the potential of presymptomatic detection of the plant diseases was demonstrated. Depending on the type and stage of disease the classification accuracy was between 65% and 90%.     

Quantum yield of non-regulated energy dissipation in PSII (Y(NO)) for early detection of leaf rust (Puccinia triticina) infection in susceptible and resistant wheat (Triticum aestivum L.) cultivars

The development and optimization of protocols for the precise and pre-symptomatic detection of diseases, and non-invasive evaluation of genotype-specific pathogen resistance enabling selection of the more promising genotypes in breeding programmes are important and often overlooked topics in precision agriculture. The increasing pressure to minimize both production costs and the environmental impact of pesticides forces the search for rapid and objective methods of screening pathogen resistance. Using the non-destructive pulse amplitude modulated (PAM) chlorophyll fluorescence imaging technique, we hypothesized that not only disease detection but also discrimination between differences in the level of resistance of wheat cultivars to the leaf rust (Puccinia triticina Erics.) pathogen can be achieved. Experiments were conducted using the cultivars Dekan and Retro as representatives of a susceptible and a highly resistant genotype, respectively. Fluorescence measurements were carried out daily on the control and on plants inoculated with P. triticina until the first small red-brown pustules appeared in the centre of chlorotic spots. In response to pathogen inoculation, the fluorescence readings showed an early characteristic increase in Y(NO) in both resistant and susceptible cultivars. The susceptible cultivar, however, showed a more pronounced difference between Y(NO) values measured on the control and inoculated leaves as well as a distinct evolution over time. Accordingly, our results indicate that Y(NO) might be suitable for discriminating between wheat genotypes as early as 2 days after inoculation. Thus, the proposed protocol might be adopted as an additional tool for the early screening of new genotypes, especially in breeding programs that aim for high resistance to disease and low crop variability for precision agriculture. However, its implementation in experimental field plots requires improvement of the measurement system and establishment of appropriate algorithms for disease pattern recognition and data analysis.     

Early pathogen detection under different water status and the assessment of spray application in vineyards through the use of thermal imagery

Remote detection using thermal imagery has potential for use in the pre-symptomatic diagnosis of abiotic stress or of early disease detection. The latter is an issue of great importance since late detection of fungus attacks or poor spray coverage are major factors contributing to weak disease control affecting fruit quality or reducing yield in grapes. In greenhouse experiments the effects on spatial and temporal variability of leaf temperature of grapevine (Vitis vinifera L. cv. Riesling) leaves inoculated with a fungal pathogen (Plasmopara viticola (Berk. & Curt. Ex de Bary) were studied in either well-irrigated or non-irrigated potted plants. Due to the high sensitivity of leaf temperature to the amount of water transpired, infra-red thermography can be used to monitor irregularities in temperature at an early stage of pathogen development. Evidence for characteristic thermal responses in grapevines was apparent well before visible symptoms appeared. Contrasting thermal effects due to the pathogen attack were found between measurements on well-irrigated and water-stressed plants. Furthermore, from a technical point of view, thermal imagery has the potential to assess the evenness of spray coverage within a canopy, hence optimizing pesticide application efficiency.     

潜伏期柑橘黄龙病宿主糖代谢及近红外光谱特征

柑橘黄龙病(Citrus Huanglongbing)是柑橘最具有毁灭性的病害,传染性极强,严重影响柑橘产业的健康发展。及时发现潜伏期未显症的病树并挖除,能够更加有效地阻碍黄龙病的传播。为此,特研究潜伏期染病叶片的糖代谢以及叶片组织光学特性的变化规律。结果表明,淀粉、蔗糖、葡萄糖和果糖在染病未显症叶片中已经出现了异常累积,分别是健康叶片的3.58、2.16、3.41、1.70倍,同时,近红外光谱的反射率出现了升高的趋势。进一步采用Random Frog算法选择前6个敏感波段(1 015、1 331、1 065、1 334、1 022、951 nm),并结合Naïve Bayes(NB)模型对潜伏期未显症柑橘黄龙病叶片进行判别,得到了97.5%的分类正确率,对阳性样本的漏检率为0,说明采用近红外高光谱技术能够实现对未显症柑橘黄龙病的检测,可为田间柑橘黄龙病的快速普查提供新的方法。     

Detection of common defects on oranges using hyperspectral reflectance imaging

To detect various common defects on oranges, a hyperspectral imaging system has been built for acquiring reflectance images from orange samples in the spectral region between 400 and 1000 nm. Oranges with insect damage, wind scarring, thrips scarring, scale infestation, canker spot, copper burn, phytotoxicity, heterochromatic stripe, and normal surface were studied. Hyperspectral images of samples were evaluated using principal component analysis (PCA) with the goal of selecting several wavelengths that could potentially be used in an in-line multispectral imaging system. The third principal component images using six wavelengths (630, 691, 769, 786, 810 and 875 nm) in the visible spectral (VIS) and near-infrared (NIR) regions, or the second principal component images using two wavelengths (691 and 769 nm) in VIS region gave better identification results under investigation. However, the stem-ends were easily confused with defective areas. In order to solve this problem, representative regions of interest (ROIs) reflectance spectra of samples with different types of skin conditions were visually analyzed. The researches revealed that a two-band ratio (R875/R691) image could be used to differentiate stem-ends from defects effectively. Finally, the detection algorithm of defects was developed based on PCA and band ratio coupled with a simple thresholding method. For the investigated independent test samples, accuracies of 91.5% and 93.7% with no false positives were achieved for both sets of selected wavelengths using proposed method, respectively. The disadvantage of this algorithm is that it could not discriminate between different types of defects.     

Early detection of Fusarium infection in wheat using hyper-spectral imaging

Infections of wheat, rye, oat and barley by Fusarium ssp. are serious problems worldwide due to the mycotoxins, potentially produced by the fungi. In 2005, limit values were issued by the EU commission to avoid health risks by mycotoxins, both for humans and animals. This increased the need to develop tools for early detection of infections. Occurrence of Fusarium-caused head blight disease can be detected by spectral analysis (400-1000 nm) before harvest. With this information, farmers could recognize Fusarium contaminations. They could, therefore, harvest the grains separately and supply it to other utilizations, if applicable. In the present study, wheat plants were analyzed using a hyper-spectral imaging system under laboratory conditions. Principal component analysis (PCA) was applied to differentiate spectra of diseased and healthy ear tissues in the wavelength ranges of 500-533 nm, 560-675 nm, 682-733 nm and 927-931 nm, respectively. Head blight could be successfully recognized during the development stages (BBCH-stages) 71-85. However, the best time for disease determination was at the beginning of medium milk stage (BBCH 75). Just after start of flowering (BBCH 65) and, again, in the fully ripe stage (BBCH 89), distinction by spectral analysis is impossible. With the imaging analysis method ‘Spectral Angle Mapper’ (SAM) the degree of disease was correctly classified (87%) considering an error of visual rating of 10%. However, SAM is time-consuming. It involves both the analysis of all spectral bands and the setup of reference spectra for classification. The application of specific spectral sub-ranges is a very promising alternative. The derived head blight index (HBI), which uses spectral differences in the ranges of 665-675 nm and 550-560 nm, can be a suitable outdoor classification method for the recognition of head blight. In these experiments, mean hit rates were 67% during the whole study period (BBCH 65-89). However, if only the optimal classification time is considered, the accuracy of detection can be largely increased.     

Weed Detection Using Canopy Reflection

For site-specific application of herbicides, automatic detection and evaluation of weeds is desirable. Since reflectance of crop, weeds and soil differs in the visual and near infrared wavelengths, there is potential for using reflection measurements at different wavelengths to distinguish between them. Reflectance spectra of crop and weed canopies were used to evaluate the possibilities of weed detection with reflection measurements in laboratory circumstances. Sugarbeet and maize and 7 weed species were included in the measurements. Classification into crop and weeds was possible in laboratory tests, using a limited number of wavelength band ratios. Crop and weed spectra could be separated with more than 97% correct classification. Field measurements of crop and weed reflection were conducted for testing spectral weed detection. Canopy reflection was measured with a line spectrograph in the wavelength range from 480 to 820 nm (visual to near infrared) with ambient light. The discriminant model uses a limited number of narrow wavelength bands. Over 90% of crop and weed spectra can be identified correctly, when the discriminant model is specific to the prevailing light conditions.     

Temporal and spatial changes of chlorophyll fluorescence as a basis for early and precise detection of leaf rust and powdery mildew infections in wheat leaves

Temporal and spatial changes in parameters of fast chlorophyll fluorescence kinetics (ground fluorescence, Fo and maximal fluorescence, Fm) and red/NIR reflectance were assessed with a Pulse-Amplitude-Modulated (PAM)-Imaging system on a daily basis over a period of 2 weeks following inoculation of wheat leaves with powdery mildew and leaf rust. The early detection of these infections by means of fluorescence imaging was possible 2–3 days before visual symptoms or significant changes in normalised-differenced-vegetation index (NDVI) became apparent. The initial infection of both fungi caused an increase in Fo and decrease in photochemical efficiency (Fv/Fm, Fv/Fo). The appearance and development of fungal pustules was accompanied by reduction in Fo and Fm. This resulted mainly from lower absorption of fluorescence exciting light by the leaf mesophyll due to the shielding effect of fungal mycelium, and to lesser extent from the chlorophyll breakdown underneath pustules. Among the evaluated fluorescence parameters, Fv/Fo displayed the most pronounced response to both kinds of infection. Mildew infection influenced chlorophyll fluorescence neither in the direct vicinity of mycelium nor in the apparently healthy leaf regions. Rust infected plants, in contrast, displayed significantly reduced photochemical efficiency Fv/Fm and Fv/Fo in chlorotic tissue around pustules. The same, but less pronounced tendency was found in the apparently healthy regions of rust infected leaves in the last days of the experiment. Dark adaptation of leaves proved to be necessary for accurate detection of both pathogen infections by means of fluorescence imaging. Additional experiments are needed to estimate the potential of this technique for remote sensing under field conditions.     

Classification of sugar beet and volunteer potato reflection spectra with a neural network and statistical discriminant analysis to select discriminative wavelengths

The objectives of this study were to determine the reflectance properties of volunteer potato and sugar beet and to assess the potential of separating sugar beet and volunteer potato at different fields and in different years, using spectral reflectance characteristics. With the ImspectorMobile, vegetation reflection spectra were successfully repeatedly gathered in two fields, on seven days in 2 years that resulted in 11 datasets. Both in the visible and in the near-infrared reflection region, combinations of wavelengths were responsible for discrimination between sugar beet and volunteer potato plants. Two feature selection methods, discriminant analysis (DA) and neural network (NN), succeeded in selecting sets of discriminative wavebands, both for the range of 450-900 and 900-1650 nm. First, 10 optimal wavebands were selected for each of the 11 available datasets individually. Second, by calculating the discriminative power of each selected waveband, 10 fixed wavebands were selected for all 11 datasets analyses. Third, 3 fixed wavebands were determined for all 11 datasets. These three wavebands were chosen because these had been selected by both DA and NN and were for sensor 1: 450, 765, and 855 nm and for sensor 2: 900, 1440, and 1530 nm. With the resulting three sets of wavebands, classifications were performed with a DA, a neural network with 1 hidden neuron (NN1) and a neural network with two hidden neurons (NN2). The maximum classification performance was obtained with the near-infrared sensor coupled to the NN2 method with an optimal adapted set of 10 wavebands, where the percentages were 100 ± 0.1 and 1 ± 1.3% for true negative (TN) classified volunteer potato plants and false negative (FN) classified sugar beet plants respectively. In general the NN2 method gave the best classification results, followed by DA and finally the NN1 method. When the optimal adapted waveband sets were generalized to a set of 10 fixed wavebands, the classification results were still at a reasonable level of a performance at 87% TN and 1% FN for the NN2 classification method. However, when a further reduction and generalization was made to 3 fixed wavebands, the classification results were poor with a minimum performance of 69% TN and 3% FN for the NN2 classification method. So, these results indicate that for the best classification results it is required that the sensor and classification system adapt to the specific field situation, to optimally discriminate between volunteer potato and sugar beet pixel spectra.     

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%).     

The potential of automatic methods of classification to identify leaf diseases from multispectral images

Three methods of automatic classification of leaf diseases are described based on high-resolution multispectral stereo images. Leaf diseases are economically important as they can cause a loss of yield. Early and reliable detection of leaf diseases has important practical relevance, especially in the context of precision agriculture for localized treatment with fungicides. We took stereo images of single sugar beet leaves with two cameras (RGB and multispectral) in a laboratory under well controlled illumination conditions. The leaves were either healthy or infected with the leaf spot pathogen Cercospora beticola or the rust fungus Uromyces betae. To fuse information from the two sensors, we generated 3-D models of the leaves. We discuss the potential of two pixelwise methods of classification: k-nearest neighbour and an adaptive Bayes classification with minimum risk assuming a Gaussian mixture model. The medians of pixelwise classification rates achieved in our experiments are 91% for Cercospora beticola and 86% for Uromyces betae. In addition, we investigated the potential of contextual classification with the so called conditional random field method, which seemed to eliminate the typical errors of pixelwise classification.     

基于光谱信息的柑橘黄龙病无损检测及分级模型构建

柑橘黄龙病(Huanglongbing,HLB)是世界柑橘生产上最具毁灭性的病害,给果农和相关产业造成了巨大的损失.以柑橘叶片为载体,利用高光谱图像技术采集柑橘叶片表面的高光谱图像,用ENVI4.7进行图像处理,提取感兴趣区域(Region of Intest,ROI),统计感兴趣区域平均光谱数据,并进行相关植被植物的运算,最后通过PLS-DA(Partial Least Squares Discrimination Analysis)判别法进行鉴别并分类.结果表明:基于平均光谱值和植被指数的PLS-DA判别模型都能对健康、缺锌和HLB叶片进行鉴别.其中基于平均光谱值的PLS-DA模型鉴别健康柑橘叶片样品的灵敏度为100％,特异度为100％,准确度为100％;鉴别缺锌柑橘叶片样品的灵敏度为80.6％,特异度为91.7％,准确度为88.9％;鉴别HLB叶片的灵敏度为89.3％,特异度为88.3％,准确度为88.9％.基于植被指数的PLS-DA判别模型鉴别健康柑橘叶片样品的灵敏度为100％,特异度为100％,准确度为100％;鉴别缺锌柑橘叶片样品灵敏度为92.5％,特异度为89.3％,准确度为90.1％;鉴别HLB叶片的灵敏度为86.4％,特异度为95.3％,准确度为90.1％.识别正确率较高,说明利用高光谱进行柑橘黄龙病病情分类是可行的.     

Spectral requirements on airborne hyperspectral remote sensing data for wheat disease detection

Remote sensing approaches are of increasing importance for agricultural applications, particularly for the support of selective agricultural measures that increase the productivity of crop stands. In contrast to multi-spectral image data, hyperspectral data has been shown to be highly suitable for the detection of crop growth anomalies, since they allow a detailed examination of stress-dependent changes in certain spectral ranges. However, the entire spectrum covered by hyperspectral data is probably not needed for discrimination between healthy and stressed plants. To define an optimal sensor-based system or a data product designed for crop stress detection, it is necessary to know which spectral wavelengths are significantly affected by stress factors and which spectral resolution is needed. In this study, a single airborne hyperspectral HyMap dataset was analyzed for its potential to detect plant stress symptoms in wheat stands induced by a pathogen infection. The Bhattacharyya distance (BD) with a forward feature search strategy was used to select relevant bands for the differentiation between healthy and fungal infected stands. Two classification algorithms, i.e. spectral angle mapper (SAM) and support vector machines (SVM) were used to classify the data covering an experimental field. Thus, the original dataset as well as datasets reduced to several band combinations as selected by the feature selection approach were classified. To analyze the influence of the spectral resolution on the detection accuracy, the original dataset was additionally stepwise spectrally resampled and a feature selection was carried out on each step. It is demonstrated that just a few phenomenon-specific spectral features are sufficient to detect wheat stands infected with powdery mildew. With original spectral resolution of HyMap, the highest classification accuracy could be obtained by using only 13 spectral bands with a Kappa coefficient of 0.59 in comparison to Kappa 0.57 using all spectral bands of the HyMap sensor. The results demonstrate that even a few hyperspectral bands as well as bands with lower spectral resolution still allow an adequate detection of fungal infections in wheat. By focusing on a few relevant bands, the detection accuracy could be enhanced and thus more reliable information could be extracted which may be helpful in agricultural practice.     

AdaBoost classifiers for pecan defect classification

One of the constraints in the adoption of machine vision inspection systems for food products is low classification accuracy. This study attempts to improve pecan defect classification accuracy by using machine learning classifiers: AdaBoost and support vector machine (SVM). X-ray images of good and defective pecans, 100 each, were segmented and features were extracted. Twenty classification runs were made to adjust parameters and 300 classification runs to compare classifiers. The Real AdaBoost classifier gave average classification accuracy of 92.2% for the Reverse water flow segmentation method and 92.3% for the Twice Otsu segmentation method. The Linear SVM classifier gave average classification accuracy of 90.1% for the Reverse water flow method and 92.7% for the Twice Otsu method. Computational time for the classifiers varied by two orders of magnitude: Bayesian (10⁻⁴ s), SVM (10⁻⁵ s), and AdaBoost (10⁻⁶ s). AdaBoost classifiers improved classification accuracy by 7% when Bayesian accuracy was poor (less than 89%). The AdaBoost classifiers also adapted well to data variability and segmentation methods. A minimalist AdaBoost classifier, more suitable for real time applications, using fewer features can be built. Overall, the selected AdaBoost classifiers improved classification accuracy, reduced classification time, and performed consistently better for pecan defect classification.     

樟子松松针锈病病原菌的鉴定

1989年在黑龙江省阿城市料甸林场,用樟子松(Pinus sylvestrisvar. mongolica Litv.)松针上的春孢子向健康的黄药(Phelledendron amurenseRupr.)和紫花铁线莲(Clematis fusca var. violacea Maxim)叶片人工接种成功。用人工接种所产生的夏孢子向相同寄主植物接种后,再次发病并产生夏孢子。由此证明阿城市料甸林场樟子松松针锈病菌(Cleosporium clematidis Barcl.)的转主寄主为紫花铁线莲和黄药,病原菌种应为铁线莲鞘锈菌(Coleosporium clema(?)idis Barcl.)。     

肾茶叶枯病病原的分离与鉴定

为了确定肾茶叶枯病致病病原菌,笔者从肾茶云南产区采集的肾茶叶枯病样本中分离到1株病原菌,并对其进行了病害症状观察,病原菌分离、鉴定和病原菌生物学特性研究。结果表明,其在PDA培养基上菌落为白色,气生菌丝发达,菌落初期下部淡粉色,后期为深黄棕色,分生孢子顶胞钩状,成熟的大型分生孢子有3～5个隔膜。将病原菌离体接种到健康肾茶叶片,保湿培养数天后接种部位出现黑褐色病斑,与田间症状一致。病原菌基因组DNA经真菌rDNA-ITS通用引物ITS1/ITS4 扩增及同源性分析,病原菌与Fusarium nematophilum,Fusarium equiseti,Fusarium chlamydosporum,Fusarium longipes聚为一支,核酸序列同源性为99.40%～99.60%。结合形态特征观察、ITS序列分析及柯赫氏法则验证结果,初步确定该病原菌为镰刀菌。     

Sensor measurements revealed: Predicting the Gram-status of clinical mastitis causal pathogens

Automatic milking systems produce mastitis alert lists that report cows likely to have clinical mastitis (CM). A farmer has to check these listed cows to confirm a CM case and to start an antimicrobial treatment if necessary. In order to make a more informed decision, it would be beneficial to have information about the CM causal pathogen at the same time a cow is listed on the mastitis alert list. Therefore, this study explored whether decision-tree induction was able to predict the Gram-status of CM causal pathogens using in-line sensor measurements from automatic milking systems. Data were collected at nine Dutch dairy farms milking with automatic milking systems and included 140 bacteriological cultured CM cases with sensor measurements of electrical conductivity, colors red, green, and blue and milk yield for analyses. In total, 110 CM cases were classified as Gram-positive CM cases and 30 as Gram-negative. Stratified randomization was used to divide the data in a training set (n = 96) for model development, and a test set (n = 44) for validation. The decision tree used three variables to predict the Gram-status of the CM causal pathogen; two variables were based on electrical conductivity measurements, and one on measurements of the color blue. This decision tree had an accuracy of 90.6% and a kappa value of 0.76 based on data in the training set. When only those CM cases were considered with extreme high probability estimates for their Gram-status (either positive or negative), 74% of all records in the training set could be classified with a stratified accuracy of 97.1%. When validated, the decision tree performed poorly; accuracy dropped to 54.5% and the kappa value to −0.20. The stratified accuracy calculated for 75% of all records in the test set was 66.7%. Predicting the CM causal pathogen showed a similar poor result; the decision tree had an accuracy of 27.9% and a kappa of 0.12, based on data in the test set. Based on these results, it is concluded that decision-tree induction in conjunction with sensor information from the electrical conductivity, color, and milk yield provide insufficient discriminative power to predict the Gram-status or the CM causal pathogen itself.     

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     

稻叶瘟染病程度的可见-近红外光谱检测方法

基于可见-近红外光谱技术,并采用偏最小二乘算法对不同水稻稻叶瘟染病程度的叶片进行化学计量学分析.分别建立基于全波段、特征波段和特征波长的稻叶瘟染病程度定量检测模型.结果表明:全波段建模的叶瘟病染病程度检测正确率达到96.7%;通过偏最小二乘算法的回归系数选择5个特征波段.分别为552～558、672～682、719～726、756～768和990～998 nm,基于特征波段的模型正确率也达到了90%,说明该5个特征波段与叶瘟病染病程度有很好的相关性;基于特征波段结果,选择5个特征波长,对叶瘟病染病程度的检测正确率为80%.说明基于可见-近红外光谱技术方法具有较好的预测能力,为稻叶瘟染病程度的快速鉴别提供了一种新方法.     

基于高光谱的琯溪蜜柚叶片磷素含量估算模型研究

蜜柚叶片磷素(phosphorus,P)含量是准确诊断和定量评价生长状况的重要指标,为快速、无损、精确地估测磷素含量,需要建立蜜柚叶片磷素含量高光谱估算模型。基于蜜柚叶片高光谱数据和磷素含量实测数据,提取原始光谱及一阶微分光谱特征波段和光谱特征变量,构建单变量估算模型、偏最小二乘回归模型和BP神经网络回归模型,并确定蜜柚叶片磷素含量最佳估算模型。在350～1 050 nm波段,原始光谱和一阶微分光谱与叶片磷素含量在可见光范围内有多波段相关性显著,并出现多个极值。原始光谱敏感波长为549和718 nm,一阶微分的敏感波长为528、703和591 nm。在建立的回归模型中,选择决定系数较高的模型进行精度检验,其中BP神经网络模型的拟合R²(0.775 9)最大,偏最小二乘估算模型的拟合R²(0.749 9)次之。综合建模精度和模型检验精度,确定BP神经网络模型为蜜柚叶片磷含量的最佳估算模型,建模和验证的R²分别为0.71和0.775 9;其次为偏最小二乘估算模型,建模和验证的R²分别为0.64和0.74...