Design And Key Technology Research Of Fish Meal Freshness Detection System Based On Bionic Olfaction

Fish meal is a kind of high-protein feedstuff that uses one or more kinds of fish as raw materials and is processed by deoiling,dehydration and comminution.Because fish meal is rich in protein and fat,in the process of transportation and storage,it is easy to produce protein decomposition,fat oxidation and rancidity,and it is easy to be polluted by microorganisms and other chemicals,resulting in the decrease of its nutritional composition,thermal energy content and palatability.Freshness can measure the degree of protein spoilage and fat rancidity of fish meal,so it is necessary to detect the freshness of fish meal.At present,artificial senses and chemical methods are used to detect the freshness of fish meal.These methods are characterized by tedious operation,labor and material consumption,and lack of objectivity.In this project,fish meal is taken as the research object,and a fish meal freshness detection system based on bionic olfactory technology is used to detect and analyze the freshness index of fish meal.This method can enrich the rapid detection methods and devices of fish meal and improve the detection level.The main research contents and results are as follows:(1)Using bionic olfactory technology,a fish meal freshness detection system is designed.The system takes the microprocessor raspberry pie as the core,which is mainly composed of gas collection,circuit data acquisition,gas sensor array,software and so on.Gas collection is mainly to collect zero gas and sample gas;circuit data acquisition is mainly to collect,amplify and AD(analog digital)conversion the electrical signal of the gas sensors;the software part designs the interface of sensor data collection,cleaning and so on,and displays the collected data in real time.(2)Using the method of single factor and response surface,taking the dispersion ratio as the test index,the sample weight,headspace space,gas flow rate,sampling time and cleaning time were optimized.Through single factor test,the headspace space is 250ml and the sample weight is 80 g.The design expert software is used to optimize these factors including the gas flow,sampling time and cleaning time,which have a great impact on the results.The optimal parameter combination is obtained.Under the optimal parameters,the feasibility and detection performance of the detection system for the quality of fish meal are verified.The experimental results show that the gas flow rate,sampling time and cleaning time all have significant influence on the dispersion ratio,and the primary and secondary order of factors is cleaning time,sampling time,gas flow rate.Considering all factors,the best combination of parameters is gas flow rate of 2.2 L/min,sampling time of 39 s and cleaning time of 77 s.at this time,the measured value of dispersion ratio is 0.579 9.Under the optimal parameters,fish meal samples with different storage time were tested,and the accuracy of different storage time discrimination was 89.4%.(3)The effect of different eigenvalues on fish meal freshness recognition was studied,and the eigenvalues of sensor array were extracted and optimized.The integral value,wavelet energy value,maximum gradient value,average differential value,relation steady-state response average value and variance value were selected as six different characteristic parameters to study fish meal samples with different storage time grades.The five recognition models which included the multilayer perceptron neural network,random forest,k nearest neighbor algorithm,support vector machine algorithm,Bayesian classification method were employed for the classification.The result showed that the RF classification method had the highest accuracy rate for the classification algorithm.The highest accuracy rate for distinguishing fish meal samples with different freshness was achieved by using the integral value,relation steady-state response average value and average differential value.The lowest accuracy rate for distinguishing fish meal samples with different freshness was achieved by using the maximum gradient value.However,when all features are used for classification,the accuracy is greatly improved.So,the all sensor features(10×6)were extracted to form the original feature matrix,and then compactness was taken as the standard to evaluate the rationality of the feature selection methods,three single feature ranking methods(MIC,?~2,F-test),three multi-feature ranking methods(RF,LR,SVM)and four recursive feature elimination methods(RFRFE,SVMRFE,DTRFE,LRRFE)were selected to carry out classification accuracy tests on fish meal with different quality.The results show that random forest-based recursive feature elimination algorithm(RFRFE)was relatively more compact and the best classification accuracy was 98.3%,at this time,the number of features was 33.After optimization,the sensor array had changed from the original 10 to 8,and the numbers of features have been reduced by 45%.In order to avoid selection bias,the 10 fold cross validation method is used,and the RFRFE algorithm was verified to be more compact again.(4)SPME-GC-MS technology was used to study the volatile gas components of fish meal,and the relationship between the volatile compounds and the response of sensor array was analyzed to verify the feasibility of the detection system.Through the detection of volatile gases 18 kinds of fish meal with different storage time,101 kinds of volatile compounds were obtained,including 11 kinds of volatile compounds such as alcohol and aldehyde.The relative contents of aldehydes,alcohols compounds decreased with the increase of storage time,the relative contents of hydrocarbons,esters and phenols compounds first increased and then decreased with the increase of storage time,while the relative contents of ketones,acids,nitrogen compounds and sulfur compounds increased.Taking the relative content of volatile compounds as the independent variable,using LDA analysis method to classify fish meal with different freshness grades,we can get that the volatile compounds must contain the overall information of the sample and the classification performance can be improved.Through the analysis of gray correlation degree between the content of various volatile compounds and characteristic markers and the response value of the sensor,we can know the selected sensor is more consistent with the volatile compounds of fish meal.Using SPME-GC-MS,the volatile components content and sensor array response data were analyzed by principal component analysis(PCA).The result of PCA of volatile components content was consistent with the result of PCA of sensor features of detection system.All of them were along the direction of PC1and PC2,and the fish meal with superior fresh,general fresh and corrupt grades was close to each other.The fish meal with fresh grade and completely corrupt grade is relatively scattered.The results showed that heptanal,benzaldehyde,(Z)-2-pentene-1-ol,1-octene-3-ol,(Z)-4-heptenal,4-methyl pentanoic acid,3,5,5-trimethyl-2-hexene,2-butanone and other volatile components contributed a lot to the freshness of fish meal,resulting in the difference of volatile odor.KNN regression method was used to establish a prediction model between various volatile components of fish meal and the response value of the sensor of the detection system.The correlation coefficient R of the test set was between 0.7734 and 0.9999,and the root mean square error(RMSE)was between0.0867-8.4655.(5)Multi linear regression(MLR),Random forest regression based on improved grid search method(IGSM-RFR),and LSSVM based on particle swarm optimization(PSO-LSSVM)were used to build the prediction model of response signal of sensor array and sample freshness index(TVB-N and AV).According to the algorithm of feature reduction and elimination based on random forest,and using the improved 10 fold cross-validation grid search to optimize the parameters of random forest regression,the optimal number of decision trees for acid value is 412,and the number of features is 21;the optimal number of decision trees for TVB-N is 233,and the number of features is 33.Then,RFRFE method is used to select the above features.Using RFR model,taking 60sensor array features and the values of temperature and humidity,62 eigenvalues in total as input and corresponding AV and TVB-N content as output,when the data of test set is input,the R~2,RMSE and RSD between the predicted AV and TVB-N content and the actual value are 0.770 4,0.465,17.49%and 0.8011,25.41,18.24%,respectively.The features of sensor array optimized by RFRFE method and the values of temperature and humidity are taken as input,and the corresponding acid value and TVB-N content are taken as output,the prediction models of acid value(AV)and volatile base nitrogen(TVB-N)with high accuracy are obtained.When the data of test set is input,the determination coefficients R~2,RMSE and RSD between the predicted value of AV and TVB-N and the real value are 0.8264,0.406,15.28%and 0.8475,22.36,16.05%,respectively.The accuracy of the model is improved by RFRFE feature selection method;Using PSO-LSSVM model,The features of sensor array optimized by RFRFE method and the values of temperature and humidity are taken as input,and the corresponding acid value and TVB-N content are taken as output,the prediction models of acid value(AV)and volatile base nitrogen(TVB-N)with highest accuracy are obtained.When the data of test set is input,the determination coefficients R~2,RMSE and RSD between the predicted value of AV and TVB-N and the real value were 0.858 5,0.362 2,13.64%and 0.878 9,20.52,14.73%,respectively.Through the comparison of three regression methods,the regression model of AV and TVB-N content established by PSO-LSSVM method has the highest precision,and can obtain more ideal prediction results.