Research On The Hyperspectral Detection Methods Of Major Components And Gelatinization Characteristics Of Millet With Different Sheep Manure Application Rates

The use of modern agricultural advanced technology to achieve efficient use of farmland nutrients,suitable for local sustainable development of regional agricultural industry is an important means for the development of organic dry farming.Shanxi Province has complex landforms and climatic characteristics,and the scale of sheep breeding and millet planting is relatively large.The combination of sheep breeding and millet planting can effectively improve the yield and quality of millet.Therefore,rapid and nondestructive testing of millet quality is of great practical significance,while traditional testing methods have the disadvantage of large investment in manpower and material resources.This technology has great application potential in realizing the hyperspectral detection of the main nutritional components and gelatinization characteristics of millet under different sheep manure application rates.In this study,the millet variety Changsheng 13 was selected as the research object,and the correlation analysis was carried out on the agronomic traits and hyperspectral characteristics of the samples by setting the application rate of 5 sheep manure.In view of the problems of time-consuming,high cost and the use of toxic chemicals in the traditional detection methods of the main nutritional components and gelatinization characteristics of millet,it is proposed to use computer software programming,hyperspectral detection technology,chemometrics,data mining,machine learning and other related knowledge The content of crude fat,crude protein,amylose,amylopectin and gelatinization characteristics of millet under the application rate of sheep manure were studied by hyperspectral detection.The main research contents are as follows:(1)In order to explore the effect of different application amounts of sheep manure on the main nutritional components and gelatinization characteristics of millet.In this paper,the agronomic traits and hyperspectral data of millet under different sheep manure application rates were analyzed to determine the optimal sheep manure application rate.And use Visual Basic 6.0 to compile hyperspectral sampling point generation,hyperspectral data preprocessing operation platform to collect and extract millet hyperspectral data,and use statistical knowledge to analyze the variation rule of millet hyperspectral reflectance.The results show that different sheep manure application rates have obvious effects on the expression of agronomic status of millet,and 6 m3 per mu treatment was the best fertilizer rate.Above and below this amount of fertilizer both have inhibitory effects on the expression of millet agronomic traits.Thousand-grain weight was significantly positively correlated with the content of amylopectin and crude protein,positively correlated with the crude fat content,and negatively correlated with the amylose content;the amylose content was extremely significantly negatively correlated with the peak viscosity and final viscosity,and The valley value viscosity was negatively correlated,and positively correlated with the gelatinization temperature;from the analysis of the variation rule of the hyperspectral reflectance of millet,it can be seen that the accumulation rule of the main nutrients of millet is different with different sheep manure application rates.The reflectivity of the 6m3 treatment was the lowest,and the CK treatment was the highest,which was in line with the expression rules of agronomic traits of millet under different sheep manure application rates.By statistical analysis,the selected three characteristic bands of 959 nm,1100 nm and 1308 nm can preliminarily draw the difference of the hyperspectral characteristics of millet under different sheep manure application rates.The hyperspectral detection of composition and gelatinization characteristics provides data support and theoretical reference.(2)In order to realize the hyperspectral detection of the main nutritional components of millet under different sheep manure application rates,the hyperspectral prediction model of millet crude fat,crude protein,amylose and amylopectin under different sheep manure application rates was established by using the hyperspectral data analysis method.The visual expression of the accumulation rules of millet crude fat,crude protein,amylose and amylopectin under different sheep manure application rates,and the Logistic-COOTBP model was proposed.The hyperspectral data analysis method adopts the feature band extraction combined algorithm to extract the spectral data feature bands,and determines that the best prediction model for crude fat and crude protein is SPA-IRIV-PLSR,and the best prediction model for amylose is MSC-RF-IRIV.-PLSR,the best prediction model for amylopectin is MSC-CARS-IRIV-PLSR,using the best prediction model to determine 4 components regression equations and visualize the accumulation of millet crude fat,crude protein,amylose and amylopectin.It can be seen from the image that the spectral characteristics were the most obvious under the treatment of 6 m3 treatment and the content accumulation rules of the main nutrient components in the reflected samples were consistent with the statistical rules of agronomic characters under the five treatments.Visual inversion can be used for rapid detection of different to provide theoretical support for the accumulation rule of the main nutrient content of millet under different sheep manure application rates;in order to improve the prediction accuracy of the amylose and amylopectin,the Logistic-COOT-BP model was proposed to analyze the amylose and amylopectin in millet powder state.The results show that the amylose using the MSC-RF-IRIV-Logistic-COOT-BP prediction model has little difference in R,RMSE,and RPD values compared with other prediction models,indicating that under different sheep manure treatments,based on the spectral data set.It was described that the amylose content changes significantly,and the prediction model has a strong linear relationship.Compared with the MSC-CARS-IRIV-PLSR prediction model for amylopectin,the model accuracy was significantly improved.The R value increased from 0.52 to 0.72,the RMSE value decreased from 7.34 to 5.25,and the RPD value increased from 1.11 to1.40.The data results show that the linear relationship of the accumulation of amylopectin content under different sheep manure treatments was not strong,and the change was relatively stable.The amount of sheep manure application has little effect,which also reflects that the variety selected in the experiment have a certain potential for stable yield and are suitable for planting in barren fields.(3)In view of the time-consuming and high cost of traditional detection methods of millet gelatinization characteristics,this paper uses hyperspectral technology combined with chemometrics and machine learning related knowledge to study the rapid detection of millet gelatinization characteristics.When millet is eaten,the gelatinization characteristics determine its food taste and commercial value.The gelatinization characteristics are mainly composed of peak viscosity(PV),trough viscosity(TV),final viscosity(FV),peak time(PT),pasting temperature(GT),break down(BD),set back(SB)7 indicators.In data processing,the spectral data of millet powder samples were preprocessed by MSC,SG,SG-MSC,CARS,and RF algorithms,and then the PLSR prediction models were established.The results show that the MSC-PLSR prediction model has the highest prediction accuracy.After the original average spectral data set was processed by MSC,the Rp of each gelatinization index can reach the highest,with PV value of 0.71,TV value of 0.75,BD value of 0.76,FV value of 0.71,SB value of 0.77,PT value of 0.82,and GT value is 0.83,which is the highest model accuracy.The RPD model rating except BD was C,other models can reach B or above,GT was A,RMSEP except PV,BD The external average energy was the lowest,and there was little difference between the CARS-PLSR and RF-PLSR prediction models and the RAW-PLSR prediction model,indicating that the use of these two key bands extraction methods can effectively reduce variables without affecting the prediction accuracy of the model.According to the characteristics of high-dimensional data,the machine learning algorithm was used to predict the gelatinization index of the sample to be tested,and BP,ELM,KELM,SVR and the sparrow search algorithm was used to optimize the data processing of the four algorithms.From the data results,the SVR prediction effect was good,which also verifies that the SVR algorithm is suitable for small sample size regression prediction.The overall fitting effect of the model without SSA optimization was poor and the accuracy was not high.The prediction accuracy of the model has been improved as a whole.Among them,the SSA optimized SVR model has the highest accuracy in predicting the FV value,with Rp was 0.9 and RMSEP was 0.06.The results show that the SSA optimization algorithm can effectively improve the model prediction accuracy.The prediction of millet flour gelatinization characteristics and the application of intelligent detection provide theoretical support.This study shows that hyperspectral technology combined with chemometrics and machine learning related knowledge can be used in the research of inversion of millet flour gelatinization index,and can provide a reference for millet quality rating,related detection sensor development and millet product deep processing.