| The storage paddy rice is often influenced by microorganism. In particular, the infection of the typical mould and toxin will lower the paddy rice quality, seriously endangering the safety of the paddy rice storage. The conventional methods of biochemical detection, such as plate culture method, thin layer chromatography method, high performance liquid chromatography, and enzyme linked immuno sorbent assay(ELISA), have certain limitations and shortcomings in detection period, technical difficulty and detection cost. Near infrared spectroscopy has the advantages of non-destructive, high analysis speed, low cost, good stability and little pollution. This method is suitable for the on-site detection, which is a potential solution.This project uses near infrared spectroscopy to achieve a rapid and nondestructive detection for the paddy rice typical storage mould and toxin. The scanning conditions of the near infrared spectroscopy analyzer and the preprocessing algorithm of the spectrum curve were studied. Moreover, this paper also studied the near infrared spectral characteristics mining of the paddy rice typical storage mould and toxin, removal algorithm of the abnormal samples and preprocessing method of the spectral data. Based on this, the mathematical model of the near infrared detection for the paddy rice typical storage mould and toxin was constructed and optimized. The qualitative and quantitative identification methods of the paddy rice typical storage mould were established. The portable index analyzer for storage paddy rice mould and toxin was designed based on the mathematical model of optimal detection, which provided a basis for a new method.In this paper, the main research issues and conclusions are obtained as follows.(1) The influencing factors of the stability of near infrared detection were discussed. The spectral data of the storage paddy rice were collected using the near infrared spectroscopy analyzer. Through the study of the near infrared transmittance spectrum response for the near infrared spectroscopy analyzer, it was found that the relative standard deviation of 12 wavelengths of the paddy rice samples at the scanning temperature of 25 ℃ was minimum, which was 3.026%. The minimum relative standard deviation was determined as 9.436% when the samples were scanned for 6 times. The spectral data of the storage paddy rice were collected using the Fourier to transform the near infrared spectrum analyzer. The changes of the near infrared transmittance spectroscopy absorbance in temperatures, resolutions and scanning times were investigated. At the scanning temperature of 25 ℃, five regions with rich spectral information existed in the paddy rice samples. The relative standard deviation at the peak was in the lowest level with 3.180%. The optimal resolution was determined as 8 cm-1 by using comparsion the mean value of the absorbance of each wave number and standard deviation. The resolution power and gathering time of the spectrum were comprehensively considered. The reasonable scanning time was determined as 64 times on the basis of the average value and standard deviation of the absorbance of each wave number. The influencing factors, such as the spectral signal-to-noise ratio and spectral acquisition time, were comprehensively considered. The results showed that the suitable scanning conditions of the near infrared spectroscopy analyzer could improve the detection performance of the transmission spectrum. The scanning conditions determined could meet the test requirements of the spectral acquisition.(2) Absorption spectral data mining of the functional characteristics of the paddy rice storage mould and toxin was explored based on the preprocessing algorithm of the spectrum curve. The raw spectra of the mould and toxin were pretreated by using the five data treatment methods, including first derivative, second derivative, multiple scattering correction, standard normal variate transformation as well as the method combining the standard normal variate transformation and the detrending algorithm. The spectral bands of characteristic absorption of the paddy rice mould and toxin were determined combined with the fundamental frequency and the double frequency absorption band of the organic functional groups in the near infrared region. According to the resolution effect of the spectral overlapping peaks after the pretreatment of the spectral curve, effects of different pretreatment algorithm on paddy rice mould and toxin spectrum were analyzed and compared. The algorithm of one derivative had the best resolution effect on the spectral overlapping peaks. The near infrared spectral bands of the characteristics of paddy rice colony, aflatoxin B1, aspergillus niger, aspergillus candidus, penicillium, aspergillus glaucus, aspergillus flavus and the mixture of five kinds of mould were determined according to the differences in the strength of the paddy rice samples at each absorption peak. The results showed that the optimum pretreatment algorithms determined could meet the specific requirements of the mining characteristic spectrum data in this test. The absorption of the paddy rice typical mould and toxin in the near infrared spectral region was mainly the absorption of the all levels of frequency doubling and sum frequency of the groups containing hydrogen, which provided the theoretical basis for the establishment and optimization of the mathematical model for the near infrared detection.(3) The mathematical model for the detection of the paddy rice mould and toxin was established by using the near infrared spectroscopy. According to the test of paddy rice storage simulation, the mathematical model of the influences of the storage environment, temperature and water content on the total number of moulds of the paddy rice surface was established. This model could describe and predict the changing rule of the total number of moulds of the paddy rice surface in the storage bin and guide the safe storage of the paddy rice. The content of aflatoxin B1 in the storage paddy rice was measured using the competitive ELISA. The studies showed that the reference data of the aflatoxin B1 content in the paddy rice samples followed the normal distribution. The effects of the removal algorithms of different abnormal samples on the predicting performance of the model was analyzed and compared, including the residual discriminant method for the concentration of sample, mahalanobis distance discriminant method, and the residual discriminant method for the leverage and students of the sample. It showed that the four algorithms chosen this paper realized the distinguishing and eliminating for the abnormal samples of the spectral data of the mould colony and aflatoxin B1. The conventional spectral pretreatment methods and wavelet analysis were used to treat the spectral data of the paddy rice aflatoxin B1. The wavelet analysis was determined as the optimal pretreatment method. Daubechies5 wavelet was chosen as the function for decomposition of wavelet basis with decomposition scale of 3. After the optimized wavelet denoising, the correlation of calibration data set and the correlation of prediction data set were 0.872 and 0.863, respectively. The standard error of calibration and standard error of prediction were 2.376 and 2.352, respectively. The prediction accuracy of the optimized wave denoising model was higher than that of the model built by the conventional spectral preprocessing method. The support vector machine regression model of the paddy rice aflatoxin B1 was established based on the algorithm of support vector machine. The optimal parameters of the model based on the RBF kernel function were determined as c =106,g =0.0015. The modeling prediction accuracy was significantly higher than that of the partial least square regression after the optimized wavelet denoising. In this study, the representative wavelengths with few points were chosen to establish the mathematical model of the near infrared prediction for the paddy rice mould colony and aflatoxin B1 based on the algorithm of the multiple linear regression. The studies showed that the algorithm of the multiple linear regression was obviously better than BP neural network algorithm. The multiple linear regression could solve the problem of the large amounts of modeling calculation of the full spectrum and the over fitting etc. In the pure inoculation condition, when the concentration of the mould was higher than 3101′ per ml, the five kinds of storage mould could be distinguished. While, for the mixed inoculation, when the concentration of the mould was higher than 5101′ per ml, the five kinds of storage mould could be distinguished. The paddy rice was in the normal storage condition, when the concentration of the paddy rice mould was 5101′ per g. In the pure inoculation condition, clustering analysis algorithm was used to realize the valid identification for the paddy rice mould type before mildew. The clustering analysis was a kind of ideal qualitative identification algorithm. When the concentration of the mould was in the range of 1101′ per ml-3101′ per ml, the linear relationship of the mathematical model established using the spectral absorption of the features and the mould concentration was significant. While, when the concentration of the mould was in the range of 1101′ per ml-6101′ per ml, the linear relationship of the established mathematical model was poor. The results showed that the near-infrared spectroscopy technology combined with the methods, such as wave analysis and clustering analysis, could effectively predict the total of the paddy mould colony and the aflatoxin B1 content. And the mould species identification and quantitative analysis of the mould content were realized under the low concentration condition. The optimized mathematical model for near infrared detection could reduce the amount of calculation and improve the accuracy of detection, which is suitable for field detection. This study can provide reference for the future portable spectrum analyzer designing applied to the field detection.(4) The portable index analyzer was designed for the field detection of storage paddy rice mould and toxin. According to the optimized mathematical model for near infrared prediction, the optical fiber probe and portable light source structure were provided in this paper. And the software system matching with the analyzer was developed, which improve the feasibility of NIR equipment. Through the test, the results of the instrument detection and the chemical results proved to be very close, which could reach the application requirements of the field detection. And it can meet the application requirements of the field detection. The results showed that the optimized mathematical model for near infrared prediction had better practicability.In this paper, the near infrared spectroscopy method combined with the spectral curve pretreatment, wavelet analysis, multiple linear regression, support vector regression and the clustering analysis method was proposed, which could achieve real time and accurate detection for the total number of fungal colonies of paddy rice and the aflatoxin B1 content. The method proposed can also achieve the fungal species identification and the quantitative analysis of the mold content under the condition of low concentration. The method is more convenient and objective than the conventional detection mode. The mathematical model of the near infrared detection proposed can overcome the deficiencies in the large amount of calculation for the full spectrum modeling and the over fitting, and improve the detection accuracy, which is more suitable for the field detection. It can provide theoretical and technical basis for the rapid, nondestructive and on-site detection for the paddy rice molds and toxins, and can also provide a new idea for quality control in the process of the paddy rice storage. |
PDF Full Text Request