Study On Non-destructive Detection Of Tan-sheep Tenderness Based On Spectral Image Information Fusion

Tenderness is a key sensory feature of meat quality directly affecting the edible value and commercial value of meat.Traditional sensory evaluation and physicochemicaldetection method are hard to realize the rapid on-line detection, single sensor nondestructive technology can hardly evaluate comprehensively and accurately. In this paper, choose the characteristics agricultural products of Tan-sheep in ningxia as the research object and collectspectral information and image information of lamb surface by hyperspectral imaging system.fusing spectral feature information and image feature information to establish a quantitative prediction model and tendernss grade discrimination model by multi-source information fusion technology implementing fast nondestructive detection and evaluation of Tan-sheep quality.The. main research conclusions are as follows:(1)Tan-sheep tenderness detection based on spectral dataPreprocess 400-1000 nm and 1000-1700nm raw spectral data of Tan-sheep and establish the PLSR model and contrast the model result picking out optimum pretreatment method The results show that 900-1700 nm spectral data modeling is not ideal; model of 400-1000 nm spectral data processed by S-G smoothing is the optimal in which Rc2= 0.79, RMSEC=0.90, Rp2= 0.72, RMSEC= 0.95. 444,487,564,588,636,684,756,919 and 953 nm are extracted as characteristic wavelength by weighting coefficient method of PLSR. establish PLSR model and LDA model of lamb tenderness using the full spectra and characteristic spectra, respectively.Rp of the full and feature spectra PLSR model are 0.85and 0.84, RMSEP are0.95; band NER of the full and feature spectra LDA model are 0.72, the effect of feature spectra information model and full spectra information model are alike, the feature spectra information is valid.(2)Tan-sheep tenderness detection based on image dataExtract eight parameters of lamb intramuscular fat distribution characteristics variables:the total fat density, larger particles fat density, medium particle fat density, small fat particles density, total particle fat concentration,large particle fat concentration, medium particle fat concentration, small particle fat concentration.Four effective characteristic variables of total fat density,medium particle fat density,total particle fat concentration,small particle fat concentration are extract by stepwise regression method further.MLR mode and LDA modell are established using effective feature variables, MLR model Rp= 0.69, RMSEP=1.10, the LDA model prediction NER=0.84.single signal source of spectra information is better than image information for tenderness prediction model; single signal source of image information is better than spectra information for tenderness LDA model;establishing lamb tenderness quantitative prediction and discriminant model using fat distribution characteristics extracted by image processing methods is feasible.(3)Tan-sheep tenderness detection based on fusion of image information data and spectra information dataDesign fusion layer of spectral and image characteristics data of Tan-sheep and establish PLSR model and the LDA model of tenderness.Results show that the feature layer integration information model measuring more accurately and Solidly than single signal source of spectra information or image information model. the fusion information model RP=0.89, RMSEP=0.73. Fusion of spectra analysis, image processing and information fusion algorithm combined with visible near infrared hyperspectral imaging techonlogy method to evaluate Tan-lamb tenderness comprehensively is feasible.