Research On Optimization Detection Technology Of Spectral Analysis And Spectral Image Fusion Of Main Fatty Acids In Lamb Meat

The flavor property and nutritional value of lamb meat were closely related to the composition and content of fatty acids.Traditional methods of physical and chemical detection of fatty acids were time-consuming and labor-intensive,which are difficult to meet the rapid and accurate detection requirements of modern meat industry.As a fast detection method,hyperspectral imaging technology can obtain the internal component information and external attribute characteristics of samples simultaneously.Therefore,in this study,the near-infrared hyperspectral imaging technique was used to rapidly detect the contents of oleic acid,palmitic acid,stearic acid and linoleic acid in lamb meat.A fatty acid quantitative detection model based on different signal characteristics(spectral information,spectral image fusion information)were proposed,the characteristic wavelength spectral interference factors of specific compound and wavelength migration effect would be explored.The response mechanism of spectral absorption of the fatty acid was revealed.Construction of a rapid quantitative detection and localization analysis system for fatty acid content in lamb meat,which provide a theoretical basis for the rapid detection of meat quality and the development of on-line detection equipment.Specific research contents and results are as follows:(1)Near-infrared hyperspectral imaging and chemometric methods were used to preprocess and analyze the spectral data of oleic acid,palmitic acid,stearic acid and linoleic acid.The results showed that the best pretreatment method for oleic acid content was the second derivative(SD)method,and the models Rc2 and Rp2 were 0.8351 and 0.8679 respectively.Orthogonal signal correction(OSC)method can remove the irrelevant information in original spectrum of palmitic acid content,its Rc2 was 0.9017 and Rp2 was 0.8633.After stearic acid content was pretreated by standard normal transformation(SNV)method and detrending(DT)method,it reduced the light scattering and other phenomena,and the model Rc2 and Rp2 were 0.8162 and 0.7574 respectively.The baseline calibration method was the optimal pretreatment method for the linoleic acid content of lamb meat,its Rc2 was 0.5767 and Rp2 was 0.4144.The full-band spectral data has a better prediction effect on the content of oleic acid,palmitic acid and stearic acid,but it has a poor prediction effect on the content of linoleic acid.(2)Four characteristic wavelength extraction algorithms were used to extract the characteristic wavelength of the spectral data after optimal preprocessing.The quantitative analysis models of oleic acid,palmitic acid,stearic acid and linoleic acid contents based on linear partial least squares regression(PLSR)and nonlinear least squares support vector machine(LS-SVM)were established respectively.The results showed that for the oleic acid content,the PLSR model established by the 18 feature variables extracted by the CARS method had the best prediction effect,the optimal prediction model was CARS-PLSR(Rc2=0.8233,Rp2=0.8750);For palmitic acid content,the LS-SVM model with 5 feature wavelengths extracted by the SPA had the best prediction effect,with Rc2 of 0.8955 and Rp2 of 0.9224.The PLSR model based on 23 characteristic variables extracted by variable combination cluster analysy-genetic algorithm(VCPA-GA)has the best prediction effect on stearic acid content,the optimal prediction model had Rc2 and Rp2 of 0.7994 and 0.8743 respectively.For linoleic acid content,the LS-SVM model established by the 25 characteristic wavelengths extracted by CARS method is the best prediction model(Rc2=0.6061,Rp2=0.5437).Therefore,the characteristic wavelength spectral data can better replace the full-band information for modeling analysis.(3)Use the gray level co-occurrence matrix(GLCM)method to sequentially extract texture information such as energy,entropy,homogeneity,and correlation of the first principal component image of lamb samples in the directions of 0°,45°,90° and 135°,the fast detection and distribution visualization of oleic acid,palmitic acid,stearic acid and linoleic acid were realized by combining the feature spectral information with the image texture information.The results showed that compared with the optimal feature wavelength data model,after the texture information was fused,the oleic acid content of PLSR prediction model Rc2 and Rp2 were increased by 0.0056 and 0.0027 respectively.The LS-SVM prediction model for palmitic acid content showed that its Rc2 and Rp2 were 0.8859 and 0.9502,respectively,higher than the single spectral data model.The content of stearic acid in LS-SVM model Rc2 increased by 0.0722.The content of linoleic acid in PLSR prediction model Rc2 and Rp2 increased by 0.0761 and 0.1090 respectively,indicating that the prediction performance of the model based on fusion data was higher than that the model based on single spectral information.Finally,the predicted results are inverted to the sample image to generate a visual distribution map,which provides a reference for the research of hyperspectral image technology in meat quality.