Study On Water-injected Mutton Detection Based On Information Fusion Of Hyperspectral And Digital Images

Moisture content dominates the quality of meat products,and its detection method is mainly non-destructive testing.In this paper,the Inner Mongolia special agricultural product Ximeng mutton was used as the research object.Study on the moisture content of mutton based on hyperspectral technology and digital image technology,the research progress of the moisture content of agricultural products at home and abroad was analyzed,the hyperspectral data and image date of the mutton samples were comprehensively studied.Based on the image data characteristics,two information fusion for mutton moisture content detection models were established.The main research work and achievements are as follows:1.Collection of mutton sample dataA test of 126 lamb samples under different water injection gradients was prepared.The Specim hyperspectral instrument was used to collect the spectral data of the lamb samples,and the Canon digital camera was used to collect the digital image data of the lamb samples.Moisture content data of actual mutton samples were determined.89 sets of samples were selected as the modeling set and 37 sets of samples were used as the prediction set to establish the moisture content detection model of the mutton sample.2.Detection of mutton moisture content based on spectral dataThe original spectral data were pretreated,and the pretreatment method was optimized by comparing the effect of the model.Results show that the pretreatment model of original spectral data by standard normal combined with multivariate scattering method has the best effect,the correlation coefficient of the model is Rc=0.7415,SEC=0.0624,RP=0.7384,SEP=0.0581Six characteristic wavelengths were extracted by PLS weighting method,the full spectrum wavelength and characteristic spectral wavelength were used to establish a least squares(PLSR)prediction model for mutton moisture content.The Rc2 and RP2 of characteristic band are higher than those of full band prediction model,and the SEC and SEP of characteristic band are lower than those of full band prediction model.It shows that the spectral information of characteristic band can replace the full band spectral information.3.Detection of mutton moisture content based on image dataThe digital images of mutton samples were preprocessed and six R,G and B color feature data were extracted.Compared with the gradual multiple linear regression(SMLR)and PLSR established mutton moisture prediction model,the results show that the mutton moisture content prediction model based on PLSR has high precision,and its correlation coefficients Rc and RP are higher than those modeled by SMLRmethod,and its standard deviation SEC and SEP are lower than those modeled by SMLR.4.Detection of mutton moisture content based on spectral and image dataTThe six characteristic spectral bands and six color eigenvalue information are information fusion,and the prediction model is established by PLSR method.The prediction correlation coefficients are Rc=0.9305,Rp=0.9112,and the root mean square error is SEC=0.0381 and SEP=0.0427.The analysis shows that the prediction result of mutton moisture content model based on feature layer fusion information is accurate and high precision.