Study On Quality Detection Of Prepared Steak Based On Hyperspectral And Ultrasound Imaging Technology

Prepared steak is a delicious and convenient prepared meat product,which has been favored by more and more consumers in recent years.While the market of prepared steak is expanding,there are some quality issues,such as foreign body contamination,unclear freshness,adulterated meat,synthetic pretending to be raw and so on.The conventional detection methods have certain limitations and cannot meet the requirements of rapid and nondestructive detection.Ultrasonic imaging technology is highly sensitive,strong penetrating,cleaning and safety,and has unique advantages in foreign body detection and texture evaluation.Hyperspectral imaging technology is comprehensive,fast and accurate,and widely used in food detection.In this study,the foreign body was carried out by ultrasonic imaging technology,freshness and adulterated meat were carried out by hyperspectral imaging technology.Finally,raw and synthetic prepared steaks were identified by the two technologies respectively,and the data fusion was used to establish the identification models.The main contents and conclusions are as follows:(1)The foreign bodies in the prepared steak were detected based on ultrasonic imaging technology.First,foreign bodies of different materials and sizes were buried at different depths of the steaks and ultrasonic imaging data were collected.The image processing method was used to segment the foreign body area and the detection rate was calculated.Then,the texture feature values of the minimum external rectangle area of the foreign body were extracted by gray level co-occurrence matrix method(GLCM),and linear discriminant analysis(LDA),k-nearest neighbor(KNN),back propagation artificial neural network(BP-ANN),extreme learning machine(ELM)were established to identify the materials of foreign body.The results suggested that in the foreign body of iron sheet,glass and broken bone,the ultrasonic signal of the foreign body of iron sheet was the strongest,and that of broken bone was the weakest.The foreign body region could be segmented by image processing,and the comprehensive detection rate was 97.78%.The accuracy of ELM was the highest among the four models,with the recognition rate of correction set of 83.33%,and that of prediction set of 76.67%.(2)The indexes of the freshness of prepared steaks were quantitatively predicted based on the hyperspectral imaging technique.First,the hyperspectral data of fresh prepared steaks with different storage time were collected,and the contents of total volatile basic nitrogen(TVB-N)and thiobarbituric acid(TBA)were determined by physicochemical method.Then,1~stt Der,2~ndd Der,MC,MSC,SG,SNVT were used to pretreat spectra,and competitive adaptive reweighted sampling(CARS),variables combination population analysis(VCPA),interval random frog(iRF),iRF-CARS,iRF-VCPA were used to select the characteristic wavelengths.Finally,the partial least squares(PLS)prediction models of freshness indexes were established,and the stability of the algorithm and the effect of the combination strategy were evaluated.The results suggested that in the TVB-N content prediction process,the best pretreatment method was 1~stt Der.Compared with CARS,the stability of VCPA was better.The optimal wavelength selection method was iRF-CARS,and R_P of the corresponding model was 0.939,and RMSEP was 1.22 mg/100g.In the TBA content prediction process,the best pretreatment method was MSC.The optimal wavelength selection method was also iRF-CARS,and R_P of the corresponding model was 0.893,and RMSEP was 0.061 mg/kg.(3)The adulterated meat content of prepared steaks was quantitatively predicted based on the hyperspectral imaging technique.First,the hyperspectral data of the prepared steaks mixed with different proportions of pork and duck were collected respectively.After pretreatment of spectrum,successive projections algorithm(SPA),CARS,VCPA,iRF,iRF-SPA,iRF-CARS,iRF-VCPA were used to select the characteristic wavelengths to establish PLS prediction models of adulterated content.The results suggested that the optimal pretreatment methods for the prediction model of adulterated pork and duck were MC and SNVT.The optimal wavelength selection methods were iRF-CARS and iRF-VCPA,19 and 11 characteristic wavelengths were selected,respectively.R_P of the corresponding model was 0.966 and 0.971,and RMSEP was 2.11%and 2.23%,respectively.(4)Raw and synthetic prepared steaks were identified based on hyperspectral and ultrasound imaging technique.First,the hyperspectral and ultrasound images of raw and synthetic prepared steaks were collected respectively,and the texture feature values of images were extracted by GLCM to establish the identification models.Then,the data of the two technologies was fused and the model of fusion data was optimized by SPA,CARS and VCPA.The best models for ultrasound imaging and hyperspectral imaging were ELM and KNN,and the prediction set identification rate of the models were 90.00%and 95.00%,respectively.After data fusion,the prediction set identification rate of the best model ELM was 97.50%.Among the three variable selection methods,the identification rate of the model prediction set established by the texture variables selected by CARS and VCPA was 100.00%.In this study,part of the quality index of the prepared steak was rapidly detected by ultrasonic imaging technology and hyperspectral imaging technology combined with image processing and stoichiometry.The study can provide theoretical reference for the rapid and nondestructive detection of steaks,and provide ideas for the combination strategy of spectral variable selection algorithm and the application of ultrasonic imaging technology in food quality detection.