Study On The Evaluation Method Of Ham Sausage Quality Based On BP Neural Network And Microphotography Technology

China is the country having established huge meat industry.Ham sausage is one of the largest products among China's meat products.Now the total output of domestic ham sausages has accounted for 1/3 of the total meat production in China.In order to ensure the rights and interests of consumers,it is especially important to accurately evaluate the quality of ham.In our paper,the pork ham sold in the market with large quantities was used as a studying material.The linear regression model based on the texture analysis by a textural analyzer and sensory scores,and the BP neural network model were established to simulate the preferences of consumers and to predict the sensory quality so that human error can be avoided.Visually observable photographs were obtained by using an upright optical microscope and imaging system,based on which the total area of pixels occupied by the stained area of protein,fat and starch was measured and calculated.Based on the percentage of pixels,the method for accurately determining the protein and starch content in ham sausage was developed.At the same time,the percentage of pixels in the stained area of protein,fat and starch as a percentage of total field of view pixels and sensory score was obtained.The linear regression model and the BP neural network pass model were developed to simulate consumer preferences.The main findings are as follows:1.By establishing a model of ham intestinal texture,75% was selected as the most suitable compression ratio for determining the hardness of ordinary ham intestines,40% was the most suitable compression ratio for determining elasticity and cohesion;75% was selected as the measurement excellent and super optimal.The optimal compression ratio of ham sausage hardness is 60%,which is the most suitable compression ratio for measuring elasticity and cohesion.In the study of linear regression model to predict the sensory quality,the physical and physiologic properties of the ham sausage measured by the texture analyzer were analyzed.The results showed that there was a good linear correlation between the measured result of hardness(H)and elasticity(S)of the ham sausage and its sensory score(HS),with correlation coefficients of 0.955 and 0.912,respectively;but correlation between other five indicators and sensory scores was not good.The linear regression equation obtained was as follows: HS = 0.579H(kg)+ 1.069S(cm).If the equation is entered into the computer of the texture analyzer,the output of the HS may be achieved,which may be feasible for routine purposes.The method may also be used to distinguish between different grades of pork ham sausage.2.Prediction of sensory quality based on BP artificial neural network was also studied.The elasticity,hardness,cohesiveness and adhesiveness of ham sausages measured by texture analyzer and sensory scores a neural network was established for predicting consumer preferences.The simulation results showed that the difference between the hedonic scores predicted by the artificial neural network and that obtained by sensory test was not significant(P > 0.05),the correlation coefficient is 0.993(P(sig)= 0.00 <0.01;RMSE = 0.0676;RSD = 0.18617),the accuracy of prediction was more than 90%.This realized replacement of human sensory test by machine measurement.Cluster analysis showed that the hedonic score of ham sausages predicted by the neural network well distinguished their different grades.3.The results of microphotography showed that the percentage of pixels in the stained area of protein,fat and starch accounted for the total area of view pixels was positively correlated with their content in the ham sausage so that this method should accurately determine their content.Protein content also positively correlated with the hedonic score obtained by sensory test whereas starch content negatively correlated with the hedonic score.In the photos of ham sausages,the distribution of protein,fat and starch can be observed.The predicting result of BP neural network established on the basis of the percentage of pixels in the stained area of protein,fat and starch accounted for the total field area of view pixels was significantly different depending on different grade.Cluster analysis results showed that the percentage of pixels in the area of protein and starch staining as a percentage of the total area of view pixels and the actual hedonic score of texture can accurately distinguish between different grades of pork ham sausage.4.The result of comparing the hedonic scores predicted by multiple regression models and neural networks with that obtained by sensory test indicated that there was a significant difference between the sensory test and the multiple regression method(P <0.05).On the other hand,there was no difference between sensory test and neural network(P >0.05).Therefore,the model based on BP neural network should be suitable for simulating human sensory test.It can be applied to actual meat production to achieve rapid evaluation of texture,physical and chemical quality.