An ASABE Meeting Presentation
Paper Number: 1008570
Application of Uninformative Variable Elimination
Algorithm to Determine Organophosphorus Pesticide Concentration with Near-infrared Spectroscopy
Jingjing Chen, PhD student
China Agricultural University, College of Engineering, cjjsym@qq
Yankun Peng, Professor, Corresponding author
China Agricultural University, College of Engineering, ypeng@cau.edu
Yongyu Li, Instructor
China Agricultural University, College of Engineering, YYLI@cau.edu
Jianhu Wu, PhD student
China Agricultural University, College of Engineering, wjhu180509@sina
Jiajia Shan, Master student
China Agricultural University, College of Engineering, jia.jia.1986@163
Written for presentation at the
2010 ASABE Annual International Meeting
Sponsored by ASABE
David L. Lawrence Convention Center
Pittsburgh, Pennsylvania
June 20 – June 23, 2010
Abstract.The traditional methods of determination pesticide concentration are time-consuming, complicated, and require a lot of pretreatment processes. The objective of this research was to develop a new method for determination the pesticide concentration by using NIR spectroscopy. Org
anophosphorus pesticide (chlorpyrifos) solution was prepared by dissolving the commercial pesticide into distilled water at different concentrations, and samples were prepared by pipetting the solutions onto the filter papers and then were evaporated by vacuum drying oven. The spectra of filter paper samples were acquired in the range of 4000-10000 cm-1. Partial least squares regression (PLSR) was used to establish prediction models for predicting pesticide concentration. The uninformative variable elimination (UVE) was used for variable selection of NIR spectra data in order to develop an effective PLSR prediction model. The UVE algorithm reduced more than 70% of the variable number. Prediction results indicated that the UVE-PLSR models which multiplicative scatter correction (MSC) and standard normal variate (SNV) were used as pre-processing of spectral data were able to predict the concentration of chlorpyrifos with the correlation coefficient (R) is 0.94 for validation set, and the root mean square errors of prediction (RMSEP) is 0.36 for validation set. Keywords. NIR spectroscopy, Organphosphorus pesticide, Uninformative variable elimination.
The authors are solely responsible for the content of this technical presentation. The technical presentation does not necessarily reflect the official position of the American Society of Agricultural and Biological Engineers (ASABE), and its printing and distribution does not constitute an endorsement of views which may be expressed. Technical presentations are not subject to the formal
peer review process by ASABE editorial committees; therefore, they are not to be presented as refereed publications. Citation of this work should state that it is from an ASABE meeting paper. EXAMPLE: Author's Last Name, Initials. 2010. Title of Presentation. ASABE Paper No. 10----. St. Joseph, Mich.: ASABE. For information about securing permission to reprint or reproduce a technical presentation, please contact ASABE at or 269-429-0300 (2950 Niles Road, St. Joseph, MI 49085-9659 USA).
Introduction
Pesticides are essential for agricultural and horticultural crops production. Pesticides are commonly classified as Insecticide, fungicide, herbicide, rodenticide, etc. These pesticides act against insects, rodents, weeds which are harmful in agricultural or horticultural planting. Normally, farmers use the pesticides following the instruction written in the package. In most cases, the pesticides are mixed with water and sprayed over the plants. Basically, after spraying fruits or vegetables with pesticide, a period of 10 to 14 days is required to allow the chemical to degrade. However, the full degradation of pesticide is not always achieved. In recent years, some farmers ignored to use the pesticide correctly and rationally. In order to chase a better insecticidal effect and the economic interests, the phenomenon of using pesticide excessively, or selling the fruits or vegetables just after spraying the
pesticide in few days are not difficult to see. And the pesticides overdosing also have the potential to contaminate the soil, air, and river. It can be concluded that misusing pesticides are harmful to not only human beings but also the environment, and it is vital to control the pesticide concentration on agricultural products for maintaining public heath conditions and protecting the entire environment.
At present, there are several ways to determine the concentration of pesticide residue, such as gas chromatography (GC), high-performance liquid chromatography (HPLC), thin-layer chromatography, supercritical fluid chromatography, chromatography-mass spectrometry, capillary electrophoresis, enzyme inhibition method, immunoassay method, and bio-sensor method and so on. In all of these analysis methods, the accuracy of instrumental analysis method is best, like GC and HPLC (Gambacorta et al., 2005). But the use of instrumental analysis method for controlling all produce is not possible due to the limitations in time and labor. Normally, at least 30 minutes are needed to measure the pesticide concentration of a single sample because of the complication in the testing process. And these instrument analysis methods just can be used in laboratory for accurate analysis and statutory inspection (Luypaert et al., 2003). Biological and chemical analysis methods were developed in recent years, but there are also some flaws, such as the pre-treatments are needed and the demanding of experimental conditions.
Nowadays, there is an ongoing interest to develop safe, fast, reliable and low-cost analytical methods for the determination of pesticide residue which could avoid the use of organic solvents and reduce the contact of operator with the toxic substances. In order to achieve this goal, a new analysis method and technology should be developed. In recent years, spectroscopy based procedures is regarded as a potential method which could solve the above problems. Spectroscopy analysis methods have been widely used in chemical industry, agriculture, medicine and other areas (Peng et al., 2008, 2009; ElMasry et al., 2007). Among the optical analysis methods, near-infrared (NIR) spectroscopy is the most popular method because of its non-destructive nature, the low cost of using equipment and the fast response times (Armenta et al., 2007), and it also has been successfully applied to quality control in food (Pi et al., 2009; Leroy et al., 2003; Subbiah et al., 2008), petrochemical, pharmaceutical, clinical and biomedical and environmental sectors (Ripoll et al., 2008). There are some researches about determination of pesticide or fungicide by using near-infrared spectroscopy. Such as Saranwong and Kawano (2005) developed a system for rapid fungicide determination by using near-infrared spectroscopy; Khanmohammadi et al. (2008) used near-infrared and mid-infrared spectroscopy method to detect metribuzin concentration in soil samples, and obtained the minimum detection limit of 17mg/kg and 9mg/kg respectively; Armenta, et al. (2007) developed PLS-NIR procedure provides a non-destructive, solvent free, fast and accurate
method which allows the determination of 120 samples per hour for determination of pesticides in commercial
formulations. All of this researches show the feasibility of using NIR spectroscopy to detect trace pesticides.
Main objectives of this research were to evaluate NIR spectroscopy as a tool for determining the pesticide concentration, and use statistical algorithm to develop a satisfied prediction model. Materials and Methods
Samples
Pesticide solution:  A commercial pesticide, containing 40% chlorpyrifos (Noposion, China) was used. Chlorpyrifos is an organophosphorus pesticide, normally used in the paddy, wheat, cotton, fruit trees, and vegetables. Distilled water was prepared in order to provide the solutions with different concentrations. A total of 24 concentration levels, from 1 mg/kg to 400 mg/kg of active ingredient were diluted based on the amount of chlorpyrifos. After preparation, the solutions were kept in conical flasks and preserved in a cool place in order to prevent chemical degradation and contamination.
Figure 1. Platform for filter paper.
Filter paper samples: It is well known that the control level of pesticide does not lie at the percent level but at the 10-6 level, even 10-9 level. It is hard to obtain a satisfactory result by the use of NIR s
pectroscopy to determine the concentration of pesticide solution. The reason being that water has several strong absorption peaks in near-infrared bands; as a result it is difficult to get the information of pesticide compared to water in the solution. In order to obtain the absorption of trace chemicals, a special method to concentrate the amount of chemicals on samples was developed. Filter paper was used as substrate, water was removed from wet substrate by drying, and then the NIR measurement was performed on the dried substrate. Normal filter papers (Shuangquan, China), in 9 cm diameter were selected. First of all, every piece of 9 cm filter-papers was sheared into four pieces each 30 mm diameter by using a special mold. Then put the filter paper onto the special platform, which was made of polystyrene foam and pins (Figure 1). Each platform was almost 20 cm long and 5 cm wide, and four pieces of filter paper could be placed on. After putting the filter papers onto the platform, 200µL of pesticide solution was gently pipetted onto each filter paper (the amount of 200µL is the volume
absorbable by filter paper without any overflow). Several pieces of filter paper samples were prepared for each concentration level. A total of 99 filter paper samples were prepared.
Drying filter paper samples: The platform with filter paper samples were carefully moved into the vacuum drying oven, at room temperature for 1 hour. After drying, samples were stored into vacuum
packing bags immediately and marked with different concentrations.
Spectrum Acquisition
An Antaris FT-NIR spectrometer (Thermo Nicolet, Waltham, Massachusetts, USA), equipped with an InGaAs detector was used. The filter paper sample was placed in a specially modified sample cell. The spectra were acquired in the range of 4000 cm-1 to 10000 cm-1 at 8 cm-1 interval. For each sample, three points were chosen randomly for the NIR measurement, and 32 scans were co-added for each point. The sample was then removed, and the spectra were collected again in the same manner. Three spectra were obtained for each sample at the same state, and averaged spectra were calculated for further evaluation. To prevent the interference of water vapor in the air, the spectra of samples were acquired immediately after taking out from the vacuum packing bags.
Pre-processing Method and Data Analysis
The Matlab 7.0 software (MathWorks, USA) was used for all calculations. A total of 99 filter paper samples were divided into two groups, 75 samples were selected as calibration set; the left 24 samples in each concentration level were put into validation sample set. Partial least squares regression (PLSR) was used to develop a prediction model. Multiplicative scatter correction (MSC) a
nd standard normal variate (SNV) were used in PLSR for pre-processing of spectral data. MSC efficiently eliminates the base line drift of the spectra which in turn reflects the more detailed characteristics of the spectra, and also removes additive and/or multiplicative signal effects (Brunet et al., 2009). The main advantage of SNV is to avoid attributes in greater numeric ranges dominate those in smaller numeric ranges. The PLSR model basing on all variables of the spectra is complex, thus a special algorithm uninformative variable elimination (UVE) was used as a method for variables selection of NIR spectra data of samples in order to develop the effective PLSR prediction model for determination the concentrations of pesticide samples.
UVE is an algorithm based on the regression coefficient b of PLSR (Chen et al., 2005; Wu et al., 2009). In the PLSR-NIR prediction model, there is a relationship between X (spectral matrix) and Y (concentration matrix):
Y = X b + e(1) where b is the regression coefficient vector, e is the error vector. The following five steps were taken to get a new spectral matrix with fewer wave bands:
1. PLSR was used to develop a prediction model in the entire wave range from 4000 cm-1 to
10000 cm-1. Cross validation was applied to the calibration set. Each time, one sample was taken ou
t from the calibration set. A calibration model was established for the remaining samples and the model was then used to predict the sample left out. Thereafter, the sample was placed back into the calibration set and a second sample was taken out. The procedure was repeated until all samples have been left out once. The root mean square error of cross validation (RMSEcv) was calculated for each of all wavelength combinations.
The best principal component (PC) number with the highest Rcv (correlation coefficient of cross validation) and lowest RMSEcv value was selected.
2.    A random matrix Ra have the same number of variables with independent variable matrix
was added into spectral matrix to be a new matrix XRa.
3. Partial least squares regression (PLSR) was used again. Leave one out cross validation
was carried between the new matrix XRa and concentration matrix Y. After each step of leave one out cross validation, a regression coefficient b  was obtained.
4. Analyzing the stability of C  value which is the ratio of the mean value of vector b  and the
standard deviation of vector b :variable used in lambda
)()(C i i i b std b mean =      (2)
5. According to the absolute value of C i  to discriminate the each spectra variable is effective or
not. All effective variables were selected and put into a new independent variable matrix, and then this new matrix and Y were used to establish a new PLSR prediction model. Results and Discussion
NIR Spectra
A total of 99 filter samples’ NIR original spectra are shown in figure 2, and the spectra of samples after pre-processing with MSC are shown in figure 3. It is obviously seen that the base line drift of the spectra is reduced in the figure 3 compared to figure 2 by the application of MSC.
Figure 2. NIR transmittance spectra of filter-paper samples with different chlorpyrifos content.

版权声明：本站内容均来自互联网，仅供演示用，请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系QQ:729038198，我们将在24小时内删除。