Research On Maturation And Intelligent Recognition Of Pork Tenderloin Based On Optical Properties And Computer Vision

Cooking is very important for human beings.Most ingredients need to be heated before eating.Cooking maturity is the indicator of the end of cooking and the main goal of cooking.Therefore,cooking maturity is a fundamental issue in food science.In the early research of the research group,the maturity value theory was constructed and a universal dynamic indicator,the maturity value(M value),was further established to characterize the degree of food maturity.However,the mechanism by which people subjectively judge food maturity is still unclear.Due to the fact that vision is the only sensory organ capable of continuously experiencing the photophysical stimuli during the maturation process of food cooking,exploring the internal mechanism of human maturity judgment from a visual perspective is the most valuable.When visual judgment is mature,it is usually to observe the surface of the food.However,during the cooking process,the final mature site is located at the geometric center of the solid ingredient,which cannot be directly seen by the cooking operator.Whether the center maturity can be determined through visual changes on the food surface remains to be studied.In addition,in the previous research,computer vision was used to initially explore a mature deep learning method for intelligent recognition of pork tenderloin center.However,the relevant recognition algorithm was based on a single neural network model,and the recognition range was not tested,so the research content was incomplete.To improve this intelligent recognition method,it is necessary to compare multiple neural network models and select the appropriate model for recognition range measurement.Therefore,based on maturity value theory,food optical properties and computer vision technology,this paper conducts the following research on pork tenderloin,a typical Chinese cooking ingredient.(1)The z _E value of subjective visual maturity of pork tenderloin is 10.8℃.The objective reaction kinetic parameters z values of brightness value,red-green value,whiteness value,orientation degree,and laser spot area are 35.6℃,23.8℃,36.4℃,14.9℃,and 11.3℃,respectively.The results showed that the difference between the z _E value and the laser spot area z value was only4%,with the smallest difference among the measured optical parameters.The change in laser spot area reflects the diffuse light change that the laser returns to the surface of the sample after multiple reflections,refraction,scattering,and absorption inside the meat sample.The visual sensation of whitening caused by the diffuse reflection change is an objective stimulus source that causes subjective visual maturity judgment.(2)Taking meat sample size,heating medium temperature and surface heat transfer coefficient as cooking environment variables,a series of central maturity values(M_cvalues)of pork tenderloin were measured,and the corresponding surface maturity values(M_s values)were calculated by numerical simulation method.There was a logarithmic relationship between M_c values and M_s values of pork tenderloin obtained through statistical analysis,and the relationship coefficient was greatly affected by temperature and sample size.(3)A series of Ms sample images were collected in batches to produce 194 images in three types of data sets(immature M_s:0～250 min,mature M_s:250～550 min,over mature M_s:550～1000 min),and random forest was extracted by extracting color,texture,and singular value features as a classifier to identify and classify the surface maturity.The accuracy rates were 93.1%,90.8%,and91.5%,respectively.According to the relationship between center surface maturity values obtained in(2),A method for identifying center maturity from the surface was constructed by determining the maturity level of the center.(4)Based on the pork tenderloin image dataset with different M_c values established earlier(22708 images in total),Inception V3 model,Mobile Net-V2 model and Efficient Net-B0 model were selected to conduct recognition and classification training on the dataset through transfer learning.The classification accuracy of the three types of models was 99.4%,99.8%and 99.8%respectively,and the time spent in each iteration was 3.22s,2.13s and3.59s respectively.It was found that Mobile Net-V2 model had the best classification effect and the shortest time consumption.It is most suitable for intelligent recognition and classification of pork tenderloin center maturity.(5)Establish test sets with different exposure parameters(EV values)and samples of different thicknesses to verify the recognition range of the three types of models.When the EV values of the three types of models are lower than-2 and higher than 2,the average accuracy of the three models drops to 65.75%and 60.75%,and when the thickness exceeds 1 cm,the average accuracy of the three models drops to 80.09%.It is obtained that the mature intelligent recognition range of pork tenderloin is:the thickness of the meat slice is less than 1 cm,and the exposure parameters of shooting conditions are between-2 and 2,We have improved the intelligent recognition research methods for mature classification and expanded the application of deep learning in core cooking problems.