Research On Real-time Detection System Of Lamb Variety Discrimination Based On Hyperspectral Imaging Technology

In this paper,3 types of sheep in Ningxia were studied,with Small Tail Han sheep,Yanchi Tan sheep and beach cold hybrid sheep as research subjects.Spectral information and data of three types of sheep lamb samples were studied using visible near-infrared hyperspectral(400-1000 nm)and near-infrared hyperspectral(900-1700 nm).The discriminant models and breed discriminant results of the visible near-infrared hyperspectral(400-1000 nm)and near-infrared hyperspectral(900-1700 nm)for small-tail Han sheep,Yanchi Tan sheep and Tanhe hybrid sheep were compared.The near-infrared hyperspectral band from 900-1700 nm was selected as the real-time detection system for mutton breed discrimination.A real-time detection system for mutton variety identification based on near-infrared hyperspectral technology was built.The main work of this paper and the results of the study are as follows:(1)Spectral information of mutton of three types of sheep,Small Tail Han sheep,Yanchi Tan sheep and Beach cold hybrid sheep,were collected using visible near-infrared hyperspectral(400-1000 nm)and near-infrared hyperspectral(900-1700 nm).The raw spectral data of 3 kinds of mutton were processed by S-G,SNV,Area normalize,Max normalize,Baseline and Deresolve.With all the samples as the calibration set,the correct recognition rate of the sample correction set was the highest in the range of 400-1000 nm,and the SNV-based PLS model had the highest recognition rate of 92.14%,and the discrimination effect was the best.In the range of 900 to 1700 nm,the normal normalized PLS model had the highest correct recognition rate for the sample correction set,reaching 93.10%with the best discrimination effect.(2)240 mutton samples from Yanchi Tan sheep(classification number 1),90 mutton samples from Small Tail Han sheep(classification number 2),90 mutton samples from Tanhe hybrid sheep(classification number 3),the division of the correction set and the prediction set was carried out.The sample was divided into a calibration set and a prediction set by 3:1 according to the SPXY method,315 correction sets were used to establish the model,and 105 prediction sets were used to verify the model.Using SPA,UVE,and CARS algorithms to extract characteristic wavelengths,in the range of 400 to 1000 nm,the characteristic wavelengths for SPA,UVE,and CARS extractions were 19,34,and 48;in the 900 to 1,700 nm range,the numbers were 5,60,and 15.(3)The PLS-DA discriminant model of full spectrum,SPA,UVE and CARS was established.In the range of 400-1000nm,CARS-PLS-DA was the best discriminant model,and the correct recognition rates of its correction set and prediction set were 89.21%and 85.71%.In the range of 900 to 1700nm,UVE-PLS-DA was the best discriminant model,and the correct recognition rate of calibration set and prediction set were 91.75%and 92.38%.The discriminant models and discriminant results of visible near infrared hyperspectral(400～1000 nm)and near infrared hyperspectral(900 to 1700nm)were compared.The near-infrared hyperspectral(900～1700nm)discriminant effect on mutton varieties is better than the visible near-infrared hyperspectral(400～1000 nm)discriminant effect.Therefore,the model established by the near-infrared hyperspectral 900～1700nm band was applied to the real-time detection system of lamb breed discrimination.(4)A real-time detection system for lamb variety discrimination based on near infrared hyperspectral imaging technology was developed.The real-time detection system included hardware and software.The hardware part had completed the design of spectrum acquisition device,sample delivery device and real-time detection device.The software system had completed the design of communication,spectral data acquisition and processing,spectral data and data analysis and real-time display.Experiments proved that the real-time detection system based on near infrared hyperspectral imaging was feasible.