Research On An Agricultural Product Inspection Based On Bionic Electronic Nose

As a large agricultural country,China has a strong geographical distribution of agricultural products,which leads to the need for large scale storage and long-time transportation of agricultural products.Traditional methods of testing agricultural products are often tedious,complicated,poorly repeatable and some of the testing methods have subjective interference factors.The electronic nose,as a kind of biological smell detection method,it has the advantages of fast,accurate,economical and non-destructive,so the use of electronic nose to identify the quality of agricultural products has a greater value.Therefore,in this paper,based on the basic sensory evaluation,combined with the physical and chemical analysis,the bionic electronic nose system for agricultural products quality recognition is studied with different typical agricultural products as the research object.In this paper,several typical agricultural products,i.e.pork and scallops from fish and meat,and strawberries and broccoli from fruit and vegetables,were used for the study and validation of the gas products of agricultural products in different quality states using chromatography-mass spectrometry.This identifies volatile products such as H2 S,C2H5OH,C2H4,NH3 and CO2 as important indicators for quality testing of agricultural products.The sensor is selected according to the characteristic gases produced during the quality change of the agricultural products,and the gas collection chamber is optimised by modelling and analysing the structural and hydrodynamic characteristics of the human nasal cavity to build a bionic electronic nose detection system.A number of typical agricultural products are tested using the instrument to discuss the feasibility of the bionic electronic nose detection technology for agricultural product testing.The sensory evaluation of several typical agricultural products such as pork,scallops,strawberries and broccoli selected above,such as TVBN testing of meat by semi-micro Kjeldahl method,provides a basis for discriminating the data from the enose testing trials.Different extraction methods in the time and frequency domains are used to analyse and study the characteristic parameters of the e-nose response curve as data analysis attributes,and three pattern recognition algorithms,K-nearest neighbour(KNN),random forest(RF)and support vector machine(SVM),are used to establish an analysis model based on the e-nose for agricultural product quality detection.The recognition effect of the feature extraction method in the time domain is generally better than that of the feature extraction method in the frequency domain in the detection of independent types of agricultural products,and all three recognition algorithm models can effectively distinguish agricultural products in different quality states,with a minimum recognition rate of 88.36% and a maximum recognition rate of 96.65%.The experiments show that the electronic nose detection for the quality characteristics of agricultural products can be efficient,convenient,accurate and non-destructive,and verify the applicability of the bionic electronic nose system.The paper concludes with the optimisation of the original array of electronic noses for actual agricultural product quality detection scenarios,and the design of a mixed agricultural product detection test to further identify spoiled agricultural product species based on the determination of the presence or absence of spoilage in agricultural samples.Using the RFECV-random forest method,the sensor array is optimised by ranking the importance of different sensors according to Gini coefficients and feature importance,removing unnecessary and redundant sensors,reducing the original array of 16 sensors,to 7 sensors,while using the optimised sensor array to compare with the original array,the recognition rate increases by 1.167% on average.For real-life mixed storage of agricultural products,experiments related to the detection of mixed agricultural products were designed,using the electronic nose system to test four selected mixed agricultural products,while setting up a fresh control calibration group,and according to the application scenario,the identification targets of mixed agricultural products were divided into four cases: spoiled meat,spoiled fish,spoiled vegetables and spoiled fruit,i.e.,in determining whether there is spoiled agricultural products in the sample In other words,the types of spoiled produce are further identified on the basis of the presence of spoilage in the produce samples.The analysis and comparison between the three algorithms used in the detection of single agricultural products and the different feature extraction algorithms in the time and frequency domains were carried out to prove that the bionic electronic nose detection system still has a good classification effect on the mixed agricultural products.The recognition rate of the model using the random forest algorithm(RF)could reach 99.63%.The results show that the optimised bionic electronic nose system is able to make effective judgements on the freshness of mixed agricultural products and shows potential for practical application.