Research On Rapid Detection Methods For The Quality Of Processed Pork Based On Hyperspectral Imaging Technology

Food quality and safety are crucially related to the national economy and people's livelihood.Therefore,advanced food detecting technology is highly needed to ensure food safety and quality.Pork is the most consumable meat in China,meanwhile,China is also the world's largest pork consumer.Hence,the quality and safety of pork have always been a major concern in China.Usually,pork is stored and processed in various ways and they present different quality characteristics under different processing.The existing assessing methods for pork quality include chemical,instrumental,and sensory analysis,which are time-consuming,labor-intensive,destructive to samples,and prone to contamination.This made it difficult to meet the requirement of objective,rapid and non-destructive detection in the modern food industry.Hyperspectral imaging technology,a combination of spectroscopy and imaging technique,has emerged as a novel,fast and non-destructive detecting technology and it has been developed in the application of food quality and safety evaluation.The current research will study pork as the research object,using visible near-infrared(VIS-NIR,400–1000 nm),short-wave near-infrared(SW-NIR,1000–2000 nm)and single-shot(465–630 nm)hyperspectral/multi-spectral imaging techniques.Combined with spectral analysis methods,digital image processing,and chemometric method,the relationship between the quality changes of processed pork and variation in spectra and image features were investigated,thereby establishing a hyperspectral imaging detection method to characterize the quality of processed pork using computer programming technology.This research is carried out in the following aspects.(1)Rapid detecting approaches on the quality of pork under a single cold or freezing process were established.The variation in p H and TVC values of chilled pork stored at 4°C was not significantly changed within 6 and 4 days,respectively,which subsequently increased significantly.The p H values of frozen-thawed pork were considerably lower than that of fresh pork,and the drip loss of frozen-thawed once was less than that of twice.The best predictive models for p H and TVC values of cold pork,as well as the p H value and drip loss for frozen-thawed meat,are using VIS-NIR-SPA-PLSR.The predicted performance of the model for above classification model to differentiate frozen-thawed and fresh pork is using VIS-NIR-SPA-PCA-GLGCM-PLS-DA(CCR=97.73%).In this way,a non-destructive testing method has been established to study the related quality variation of cold or frozen processed pork.(2)Rapid detecting approaches on the quality of pork under a single boiling process were built by using different spectral pre-processing methods.The fat content in boiled pork first increased and then decreased.The ratio of saturated and unsaturated fatty acids did not change significantly.The contents of sarcoplasmic and myofibrillar protein decreased with cooking time and there were no more significant changes after cooking for 120 s onwards.The best predictive/classification models for the quality of boiled pork are provided individually(fat content:AV-original model,R_P2=0.7884,RMSEP=0.34%;saturated fatty acid content:RN-SNV model,R_P~2=0.8030,RMSEP=0.46%;unsaturated fatty acid content:RV-1st-der-COW model,R_P~2=0.7984,RMSEP=0.42%;sarcoplasmic protein content,RV-1st-der-COW model,0.9185,RMSEP=0.95%;classification on the degradation of macromolecular proteins,RCR-MSC-SPA-SVM model,CCR=88.89%).The spectral preprocessing method is of significance in the improvement of model accuracy,especially the newly introduced method of COW in this study has exhibited great application potential.(3)Samples were extended from single processed pork to various processed pork,and rapid detecting approaches on the moisture content of pork under various processing were constructed by applying different image feature extraction methods.The best predictive model for moisture wavelengths of 1019,1134,1159,1255,1446,1752 and 1911 nm.The model based on the extraction of optimal wavelengths was proven to have the best effect on the prediction of moisture content,whereas,it was less effective to use image texture features alone.Although the performance of the new method,i.e.spectral absorption index,is slightly lower than that of the optimal wavelengths,fewer features were needed to be selected.(4)Rapid detecting approaches on protein-related substances of pork under various processing were established by investigating different chemometric regression methods.There was a negative correlation between moisture and crude protein content.The best predictive models for different contents are provided individually(protein:SW-NIR-LS-SVM model,R_P~2=0.8961,RMSEP=0.82%;sarcoplasmic protein:SW-NIR-MLR model,R_P~2=0.8848,RMSEP=selection of the regression model has a great impact on the final predicting performance.(5)Rapid detecting approaches on fat-related substances of pork under various processing were built based on a combination of different data analysis methods.There was a negative correlation between moisture and crude fat content.The best predicted models for different contents are provided individually(fat:VIS-NIR-SNV-RC-PLSR model,R_P2=0.8097,RMSEP=0.58%:saturated fatty acid:VIS-NIR-MSC-UVE-SPA-PLSR model,R_P2=0.7890,RMSEP=0.78%).Collaborative analysis using multiple data processing methods has positive significance to improve the performance of predictive models.(6)On the basis of using hyperspectral imaging technology to detect the quality of various processed pork,the application of single-shot multi-spectral imaging technology has transferred hyperspectral imaging to multi-spectral imaging,which has been used to establish a multi-spectral method to detect the content of principal components of pork.When the same wavelengths were selected,the difference in prediction of pork principal components is small between linear scanning hyperspectral imaging and single-shot multi-spectral system,and the selection of the optimal wavelengths has a great impact on model performance.Meanwhile,the feasibility of miniaturization of the existing single-shot multi-spectral equipment(16wavelengths)was explored.The results showed that when nine integrated optimal wavelengths(402,408,411,414,421,435,466,626,and 991 nm)were selected to predict the content of principal components,R_P~2 was larger than 0.79 using VIS-NIR predictive model,which provided theoretical and technical support for the rapidity and miniaturization of hyperspectral imaging technology.