Study On The Multi-intelligence Fusion Technology For The Meat Freshness Identification Based On Bionic Nose

To improve the efficiency of meat freshness detection, the bionic electronic nose and tactiletechnology were studied with frozen chicken and pork samples based on the original sense testingcombining physical and chemical index. The multi-sensor information fusion technologies thatcan be used to identify meat freshness based on bionic tactile and electronic nose technologieswere essentially explored in this study.Based on the CT images of a nose of human, a3D model of nasal cavity was established. Theairflow dynamics were studied, using computational fluid dynamics (CFD), to analyze the effectsof the nasal cavity structure and airflow dynamics inside the nasal cavity on the olfactorysensation. Accordingly, a gas-chamber system of bionic olfactory sensation imitating the nasalcavity was designed. It has been verified that odor can be transported to the locations where thesensitive components of each sensor assembled in the gas-chamber system; and their contact andreaction time can meet the requirement of valid absorption time of sensors.The effects of eigenvalues on the identification of meat freshness were studied and thefeature extraction was completed for the response signal of sensor array of bionic nose. The RBFneural net, BP neural net and support vector machine (SVM) were applied for the identificationanalysis. The mean eigenvalues in all data set were very excellent in comparison with thepredicting results obtained from different eigenvalue selection methods. Comparing the predictingresults obtained from the three identification models, it shows that the SVM had the optimumidentification rate and this superiority was more obvious when the samples were lesser. Theaccuracy of the system in recognizing different meat freshness level for chicken and pork sampleswas up to92.35%and91.49%, respectively, at8℃; and,90.87%and90.48%, respectively, at0℃.The identification efficiency of SVM meat freshness using bionic nose with optimized array was91.47%and it increased3%compared with the initial array.The WDW-20J electronic universal test machine was used to simulate the tactile sensation ofhuman and obtain the meat elasticity information. The original characteristic information of bionictactile sensation was extracted using primary component analysis (PCA) and the characteristicvariables were extracted as well. The models of meat freshness identification were establishedapplying the linear identification analysis, RBF neural net and combined net combining thecharacteristic information from PCA. The results show that identification model of hereditaryoptimization RBF neural net had the optimum identification rate. Its accuracy for identifyingchicken and pork met samples was up to85.54%and85.12%, respectively, at8℃; and,83.37% and82.91%, respectively, at0℃.The original characteristics information of meat elasticity and odor “fingerprint” wereabstracted from chicken and pork samples frozen at different conditions. The multi-sensor fusionidentification models were established applying the multi-sensor information fusion technologiesand utilizing the characteristic fusion information of bionic tactile and electronic nose abstractedby SVM combing PCA. The experimental results show that identification rate of multi-sensorfusion identification model was superior to each individual identification method. Its identificationrate for chicken and pork samples was up to95.20%and93.45%, respectively, at8℃; and,94.23%and92.11%, respectively, at0℃. It experimentally verified the high feasibility andaccuracy of combining the electronic nose and tactile technology for meat freshness detection.