| Citrus,as the first type of fruit,has always been loved by people.With the change of national consumption concept in recent years,its requirements for fruit quality are increasing day by day.my country is facing the consumption dilemma of structural oversupply and short supply.The production dilemma of low standardization and uneven quality of fruit.As a key indicator of the internal quality of citrus,sugar content has the characteristics of difficulty in manual identification and low efficiency.Rapid and non-destructive detection of sugar content of citrus is of great significance to improve the economic benefits of citrus.According to the existing research on fruit detection equipment and the near-infrared spectrum detection technology used at home and abroad,there are problems in the stage of research:relatively static detection,dynamic online detection is less;sorting collaborative control system mostly adopts single chip microcomputer,which is difficult to popularize;There are few studies on the spot size that affects the detection area in online detection;there is a lack of fast and effective algorithm combination methods.According to the above problems,this study determined the corresponding technical route.The main contents and conclusions are as follows:(1)The optical path design and layout of the spectral detection device are studied,and a double-lens optical path based on the transmission type is designed,which can realize the adjustment of the spot size.The functional requirements of the detection software in the spectrum acquisition process are analyzed,and the spectrum online detection software is developed according to the requirements.According to the integrated characteristics of system detection and sorting,a control system with PLC as the control core was designed and developed,including the coordinated control of external triggering of the spectrometer.(2)The online detection conditions are analyzed.Firstly,the collection period and the detected parts of citrus in the online detection were analyzed,and the variable integration time collection mode was proposed.Spectrograms collected at different speeds,different integration times and different spot sizes were analyzed and their effects on the detection results were analyzed.When the speed increases,the production capacity of the whole machine will be greatly improved,but the detection accuracy will be greatly reduced;the increase of the integration time has a positive correlation with the quality of the modeling results,and the integration time is usually related to the production line speed;research The influence of the size of the light spot on the detection results was studied,and the results showed that the small light spot with higher illumination efficiency could obtain more ideal spectral information in a limited integration time.Therefore,5 points/s are adopted,the integration time is 100ms,and the detection conditions of small light spots are used as the reference conditions for the subsequent model establishment.(3)The online detection test was carried out with the self-developed Brix online detection system.According to the determination of the detection conditions,the corresponding spectral data were collected online for 124 citrus samples,and the Brix value was measured.The comparisons of outlier removal methods,spectral preprocessing methods and characteristic wavelength selection methods are presented.According to the modeling results,MCCV in the outlier removal method achieved the best effect,the model was 0.875,and the RMSEP was 0.426°Brix;in the spectral preprocessing method,the MSC achieved the best effect,the model was 0.872,and the RMSEP was 0.872.is 0.425°Brix;among the characteristic wavelength selection methods,SPA selects the fewest wavelength points,but the modeling effect is not very satisfactory.The wavelength points selected by CARS-SPA are medium,but only account for 3.51%of the original spectrum,and the best modeling effect.(4)For the problem of how to choose the modeling conditions,take the Vis/NIR spectral data collected online as the research object,and optimize the modeling conditions through the orthogonal experimental design,including the outlier elimination method,the preprocessing method and the characteristic wavelength selection method.The results show that,compared with the single factor selection optimization conditions,this method can cover the interaction effects between factors,select the optimal modeling conditions more accurately,and also greatly improve the selection efficiency.Modeling condition optimization provides a new approach.Under optimal conditions,the coefficient of determination and root mean square error(RMSEC)of the model calibration set were 0.961 and 0.252,respectively,and the coefficient of determination and root mean square error(RMSEP)of the prediction set were 0.904 and 0.305,respectively.A complete production line was designed,and the optimal modeling was used to conduct a comprehensive sorting performance test.When the samples were divided into four grades,the system sorting accuracy rate was 93%;when the samples were divided into three grades,the system sorting the correct rate of selection is 100%,which indicates that the system can realize the online real-time detection and sorting of the sugar content of fertile orange. |
PDF Full Text Request