Ce Yang

Analysis of soil phosphorus concentration based on Raman spectroscopy

Raman spectra signature can provide structural information based on vibrational transitions of irradiated molecules. In this work, the quantity reflecting mechanism of soil phosphorus concentration was studied based on Raman spectroscopy. 15 sand soil samples with different phosphate content were made in laboratory and the Raman signatures were measured. The relationship between sand soil Phosphorus concentration and soil Raman spectra was explored. Then the effective Raman signal was extracted from the original Raman spectra by using bior4.4 wavelet packet. The relationship between sand soil phosphorus and their extracted signals was analyzed and the PLS (Partial Least Squares) model for predicting phosphorus concentration in the soil was established and compared. The maximum accuracy model comes from the extracted effective Raman spectra after the first level decomposing. The calibration R2 was close to 1 and the validation R2 reached to 0.937. It showed high potential in soil phosphorus detecting by using Raman spectroscopy.

Hyperspectral imaging for classification of healthy and gray mold diseased tomato leaves with different infection severities

Spectral reflectance was used for classifying different types of samples.Features ranking method performed excellently in effective wavelengths selection.Early disease detection obtained the classification rate of 66.67%.Selected wavebands can be used for designing a polychromatic detection camera. This study used hyperspectral imaging technique to classify healthy and gray mold diseased tomato leaves. Hyperspectral images of diseased samples at 24h, 48h, 72h, 96h and 120h after inoculation and healthy samples were taken in the wave range of 3801023nm. A total of ten pixels from each sample were identified as the region of interest (ROI), and the mean reflectance values of ROI were calculated. The dependent variables of healthy samples were set as 0, and diseased samples were set as 1, 2, 3, 4 and 5 according to infection severities, respectively. K-nearest neighbor (KNN) and C5.0 models were built to classify the samples using the full wave band set. To reduce data volume, features ranking (FR) was used to select sensitive bands. Then, the KNN classification model was built based on just the selected bands. This later procedure of reducing spectral dimensionality and classifying infection stages was defined as FR-KNN. Performances of KNN classifier on all wave bands and FR-KNN were compared. The overall classification results in the testing sets were 61.11% for KNN, 54.17% for C5.0 and 45.83% for FR-KNN model. When differentiating infected samples from control, the testing results were 94.44%, 94.44% and 97.22% for each model, respectively. In addition, early disease detection (1dpi) obtained the results of 66.67% for KNN, 66.67% for C5.0 and 41.67% for FR-KNN. Therefore, it demonstrated that hyperspectral imaging has the potential to be used for early detection of gray mold disease on tomato leaves.

Spectral Signatures of Blueberry Fruits and Leaves

One hundred and eighty eight blueberry fruit and leaf samples were obtained from a commercial blueberry field in Waldo, Florida in June, 2010. Spectral reflectance was measured in the ultraviolet (UV), visible and near-infrared (NIR) ranges (200 nm-2500 nm) with an increment of 1 nm for each sample. Five different categories (mature fruit, intermediate fruit, immature fruit, light-green leaf and dark-green leaf) were used for classification model construction and validation. Significant differences in reflectance among the five categories were found in the visible and NIR region. Especially, the mature fruit had much lower reflectance in both regions, which shows great potential for distinguishing mature fruit from other categories. Based on the spectral characteristics of each category, fourteen normalized vegetation indices were developed for further statistical analysis to find significant bands for classifying different fruit maturity status as well as leaves. Principal component analysis (PCA), classification regression tree and multinomial logistic regression were conducted to develop prediction models for distinguishing different classes. The multinomial logistic regression model with three independent variables, which are the combinations of reflectance at six wavelengths (500, 525, 550, 575, 680, and 750 nm) performed the best, with prediction accuracy of 100%. The six wavelengths thus can be used for developing an easy-to-use and low cost fruit maturity sensor for a blueberry yield mapping system.

Band Selection Using Forward Feature Selection Algorithm for Citrus Huanglongbing Disease Detection

, 2015Purpose: This study investigated different band selection methods to classify spectrally similar data - obtained from aerial images of healthy citrus canopies and citrus greening disease (Huanglongbing or HLB) infected canopies - using small differences without unmixing endmember components and therefore without the need for an endmember library. However, large number of hyperspectral bands has high redundancy which had to be reduced through band selection. The objective, therefore, was to first select the best set of bands and then detect citrus Huanglongbing infected canopies using these bands in aerial hyperspectral images. Methods: The forward feature selection algorithm (FFSA) was chosen for band selection. The selected bands were used for identifying HLB infected pixels using various classifiers such as K nearest neighbor (KNN), support vector machine (SVM), naive Bayesian classifier (NBC), and generalized local discriminant bases (LDB). All bands were also utilized to compare results. Results: It was determined that a few well-chosen bands yielded much better results than when all bands were chosen, and brought the classification results on par with standard hyperspectral classification techniques such as spectral angle mapper (SAM) and mixture tuned matched filtering (MTMF). Median detection accuracies ranged from 66-80%, which showed great potential toward rapid detection of the disease. Conclusions: Among the methods investigated, a support vector machine classifier combined with the forward feature selection algorithm yielded the best results.Keywords: Bayesian classification, Hyperspectral, K nearest neighbor, Multi-modal Bayesian classification, Support vector machine

Hyperspectral image dataset for salt stress phenotyping of wheat

United States Department of Agriculture-Agricultural Research Service the National Science Foundation (IOS 1025881 and IOS 1361554) Minnesota Agricultural Experiment Station

A Novel Approach to Assess Salt Stress Tolerance in Wheat Using Hyperspectral Imaging

Salinity stress has significant adverse effects on crop productivity and yield. The primary goal of this study was to quantitatively rank salt tolerance in wheat using hyperspectral imaging. Four wheat lines were assayed in a hydroponic system with control and salt treatments (0 and 200 mM NaCl). Hyperspectral images were captured one day after salt application when there were no visual symptoms. Subsequent to necessary preprocessing tasks, two endmembers, each representing one of the treatment, were identified in each image using successive volume maximization. To simplify image analysis and interpretation, similarity of all pixels to the salt endmember was calculated by a technique proposed in this study, referred to as vector-wise similarity measurement. Using this approach allowed high-dimensional hyperspectral images to be reduced to one-dimensional gray-scale images while retaining all relevant information. Two methods were then utilized to analyze the gray-scale images: minimum difference of pair assignments and Bayesian method. The rankings of both methods were similar and consistent with the expected ranking obtained by conventional phenotyping experiments and historical evidence of salt tolerance. This research highlights the application of machine learning in hyperspectral image analysis for phenotyping of plants in a quantitative, interpretable, and non-invasive manner.

Detection of cold stressed maize seedlings for high throughput phenotyping using hyperspectral imagery

Hyperspectral imaging can provide hundreds of images at different wave bands covering the visible and near infrared regions, which is superior to traditional spectral and RGB techniques. Minnesota produced a lot of maize every year, while the temperature in Minnesota can change abruptly during spring. This study was carried out to use hyperspectral imaging technique to identify maize seedlings with cold stress prior to having visible phenotypes. A total of 60 samples were scanned by the hyperspectral camera at the wave range of 395-885 nm. The spectral reflectance information was extracted from the corrected hyperspectral images. By spectral reflectance information, support vector machine (SVM) classification models were established to identify the cold stressed samples. Then, the wavelengths which could play significant roles for the detection were selected using two-wavelength combination method. The classifiers were built again using the selected wavelengths. From the results, it can be found the selected wavelengths can even perform better than full wave range. The overall results indicated that hyperspectral imaging has the potential to classify cold stress symptoms in maize seedlings and thus help in selecting the corn genome lines with cold stress resistance.

Hyperspectral imaging to identify salt-tolerant wheat lines

In order to address the worldwide growing demand for food, agriculture is facing certain challenges and limitations. One of the important threats limiting crop productivity is salinity. Identifying salt tolerate varieties is crucial to mitigate the negative effects of this abiotic stress in agricultural production systems. Traditional measurement methods of this stress, such as biomass retention, are labor intensive, environmentally influenced, and often poorly correlated to salinity stress alone. In this study, hyperspectral imaging, as a non-destructive and rapid method, was utilized to expedite the process of identifying relatively the most salt tolerant line among four wheat lines including Triticum aestivum var. Kharchia, T. aestivum var. Chinese Spring, (Ae. columnaris) T. aestivum var. Chinese Spring, and (Ae. speltoides) T. aestivum var. Chinese Spring. To examine the possibility of early detection of a salt tolerant line, image acquisition was started one day after stress induction and continued on three, seven, and 12 days after adding salt. Simplex volume maximization (SiVM) method was deployed to detect superior wheat lines in response to salt stress. The results of analyzing images taken as soon as one day after salt induction revealed that Kharchia and (columnaris)Chinese Spring are the most tolerant wheat lines, while (speltoides) Chinese Spring was a moderately susceptible, and Chinese Spring was a relatively susceptible line to salt stress. These results were confirmed with the measuring biomass performed several weeks later.

Hyperspectral band selection using Kullback-Leibler divergence for blueberry fruit detection

This paper explores the feasibility of using hyperspectral imagery for blueberry fruit detection. Some bands of hyperspectral images offer redundant information. A Kullback-Leibler divergence (KLD) based band selection method is used to select the most useful bands. Forty hyperspectral images of blueberry plants were taken from the field with 1-millimeter special resolution. The proposed KLD based band selection method took advantage of the high spatial resolution of the blueberry hyperspectral images. The performance of the selected bands was compared with an unsupervised band selection strategy. The selected bands may be used for developing a blueberry detection system with a multi-spectral camera, which is of much lower cost than a hyperspectral camera.

Blueberry fruit detection by Bayesian classifier and support vector machine based on visible to near-infrared multispectral imaging

Early yield estimation based on computer vision enables better labor deployment and lower harvest expense in a large scale blueberry field. In this study, ninety multispectral images with near-infrared (NIR), red(R) and green(G) bands were collected from southern Highbush blueberry variety ‘Sweetcrisp’ from a commercial blueberry field in Waldo, Florida, during 20 April 2011 and 15 May 2011. Five thousand pure fruit pixels and 5000 background pixels were collected from the images. 66% of them were in a calibration set and the other 34% were used as a validation set. Various representations of the multispectral color models (MHSI, MYIQ, MYCbCr) originated from the NIR-R-G color model were used as the features. Bayesian classifier and support vector machine were applied for the classification of the fruit and background classes. Principle component analysis was applied before Bayesian classification for the optimized use of the features. Results show that support vector machine outperformed the Bayesian classifier with higher true positive rate (84% for fruit class and 73% for background class) and lower false positive rate (27% for fruit class and 16% for background class) in the fruit/background classification. In addition, 1000 pixels of each of eight classes (mature fruit, mid-mature fruit, young fruit, leaf, branch, soil, sky and reference board, which were found in most images) were also classified by using the two classification techniques. The true positive rates for mid-mature fruit and young fruit class were around 50%, which indicates that the color spaces were not useful for the classification of different fruit stages.