| Rapeseed is the source of main edible vegetable oil and vegetable protein for feed in China.In actual production,controlling the internal quality of rapeseed is particularly important to ensure the quality of edible oil and vegetable protein meal for feed.Aiming at the problems that traditional detection methods for oil and protein content of rapeseed damage the samples,and it takes a lot to do the detection,and the workload of manual sampling and spectrum collection during the near-infrared spectroscopy analysis of rapeseed is large,an automatic quantitative sampling with the miniature near-infrared spectrometer module which for non-destructive testing of rapeseed oil and protein content was developed,which can replace manual sampling,reduce sample loss;and improve the efficiency of rapeseed quality testing.The main contents and conclusions of the study are as follows:(1)The quality detection methods of rapeseed were summarized.Comparing the traditional detection methods with new detection methods such as near-infrared spectroscopy,hyperspectral imaging,nuclear magnetic technology and terahertz spectroscopy,near-infrared spectroscopy was determined as the detection method for the subject;the principle and steps of the near-infrared spectral analysis technology were introduced;the research status of the near-infrared spectral analysis technology and equipment in the field of rapeseed quality inspection at home and abroad was expounded;the research content of this subject was formulated and the technical roadmap was drawn based on the above theoretical research.(2)The overall design of the quantitative sampling detection device was carried out.①The basic mechanism of the device was determined which including the conveying system,the counting control system,the spectrum acquisition system,the main frame etc;the working process of the device that the conveying system transports the rapeseed samples of 1000 to the spectrum acquisition position,the spectrum acquisition system completes three spectrum acquisition of rapeseed samples in the conveying system and the counting control system controls the collection amount of the samples and the work of the control mechanism was clarified;②Refering to the rapeseed precision seeding technology;adopting the scheme of filling;carrying and gravitational sampling in the hole;designing the socket wheel sampling mechanism;expounding the composition and working principle of the sampling mechanism;and determining the diameter of the hemispherical hole of the sampling mechanism,the diameter of the seeding wheel was 80mm and the width was 20mm;the gap in the sampling mechanism was selected as 0.7mm;and the seed protection’s radian is 110°;designing the wheel-type circulation mechanism referring to the mechanical principle,determining that the number of grooves of the sheave wheel was 4,the number of round pins is 1,the center distance is 80mm;and the radius of the round pin is 6mm;③A cyclic sampling control system was designed.Determining the composition of the cyclic sampling control system including the sampling control part,the cyclic control part,the counting control part and the programmable dual-axis stepper motor controller;debugging the counting control system;and setting the sensitivity threshold that can remove impurities outside the rapeseed particles;the calculation analysis and simulation test determine that the motor speed of the sampling control part was 60r/min,and the motor speed of the cycle control part was 30r/min.(3)The establishment method of calibration model and data acquisition were analyzed.The source and processing method of test materials,parameter information of the instrument involved,spectral acquisition method and operation steps,the determination method and principle to test oil content and protein content were described;eight pretreatment combinations composed of seven pretreatment methods,such as first derivative and second derivative,were used for spectral pretreatment;mahalanobis distance algorithm was used to eliminate abnormal samples;KS and SPXY algorithms were used to divide the sample set;the optimal number of principal components was determined according to the cross validation root mean square error RMSECV value;PLS method was used to establish the calibration model.(4)The modeling method of near infrared detection of oil content and protein content was studied.① SPXY algorithm was used to divide the sample set;eight kinds of processing combinations were used for spectral pretreatment and PLS correction models of oil content and protein content were established respectively;for PLS correction models of oil content:after combination 8,the calibration model’s R_c~2 was 0.9652,R_p~2 was 0.7369;for PLS correction model of protein:the prediction accuracy of the model was improved after pretreatment;mahalanobis distance algorithm was used to remove abnormal spectra and 8 kinds of pretreatment combinations were used to re-determine the number of principal components and establish the oil content and protein content correction model;for the oil content model:after abnormal sample elimination and combination 4 pretreatment,the R_c~2 and R_p~2 value was 0.9377 and 0.7530 respectively;for the protein content model:the prediction ability of protein content correction model established after abnormal sample removal and combination 8 pretreatment was the best,R_c~2 and R_p~2 was 0.9043 and 0.7589 respectively;②A linear model PCR was selected to establish a correction model for oil content and protein content;and the results were compared with those of PLS model;for correction models of oil content:R_c~2 and R_p~2 values of PCR correction model(0.8424,0.7418)were lower than the R_c~2 and R_p~2 values of PLS correction model(0.9377,0.7530);the RMSEC and RMSEP values of PCR model(0.5803,0.4641)were greater than those of PLS correction model(0.3773,0.4442);for protein content model:the R_c~2 value(0.8146)was smaller than that of PLS model(0.9043);but R_p~2 value was larger than that of PLS;the oil content and protein content correction model in the 600-925nm band established by Flame-s was compared with the oil content and protein content correction model in the near-infrared full band established by Antaris Ⅱ;the results showed that:for the oil content model,the R_c~2 and R_p~2 values of the Antaris Ⅱ model were 0.9432 and 0.8915,which were slightly higher than those of the Flame-s model(0.9377,0.7530);for the protein content model,the R_c~2 and R_p~2 values of the Antaris Ⅱ model were 0.9233 and 0.9054,which were better than those of the Flame-s model(0.9043,0.7589).(5)The rapeseed quality detection system in the automatic quantitative sampling and detection device for rapeseed quality was integrated and developed.①The hardware part of the rapeseed quality detection system was made of Ocean Optics Flame-s micro near-infrared spectrometer,solebo Optics SLS201 stable halogen tungsten lamp which was selected by comparing different types of light sources,which can cover the effective band of the spectrometer,Ocean optics QR400-7-VIS-NIR fiber acquisition probe and independently designed sample plate for the spectrum acquisition of thousands of rapeseed;② Rapeseed quality analysis module was designed based on Matlab GUI.Rapeseed quality was comprehensively evaluated and graded when spectral data continuously collected by automatic quantitative sampling and detection device of rapeseed quality and the average thousand-grain of sample set were imported into the module;③30 rapeseed samples were continuously sampled by automatic quantitative rapeseed quality testing device for external validation;the determination coefficients of oil content and protein were 0.9224 and 0.8890 respectively;it was showed that there was small error between the predicted and actual value,while the error of protein content was a little larger.In general,the model basically meeted the requirements of device design. |
PDF Full Text Request