Research And System Design Of Microwave Transmission-based Wheat Quality Inspection Technology

Wheat is one of the most important staple foods in China,and its quality directly affects people’s lives and health.Microwave signals reflect the rich information inside the material and are of great interest in non-destructive testing of grain quality.In this paper,we propose a microwave transmittance-based double outlier hybrid detection structure to meet the needs of accuracy and speed of quality detection of wheat mold and other grains,and develop a microwave analysis device for quality detection of wheat or other grains based on this.Using wheat as the test object,we focus on the quantitative detection of aflatoxin B₁（AFB₁）and the quantitative detection of moisture content of wheat based on microwave technology.The details of the study are as follows:（1）Hardware design of the microwave device.The hardware circuit and mechanical structure are designed separately,including the hardware system composed of microcontroller,microwave source,microwave detection front-end,broadband microstrip antenna,amplitude phase detector,mixer,etc.;the mechanical structure mainly contains metal shielding cavity,carrier box,device shell and base.The hardware circuit and mechanical structure are designed separately,including the hardware system composed of microcontroller,microwave source,microwave detection front-end,broadband microstrip antenna,amplitude phase detector,mixer,etc.;the mechanical structure mainly contains metal shielding cavity,carrier box,device shell and base.Microwave device modules reasonable layout to achieve portable,integrated design,reduce the device system noise to improve detection accuracy and sensitivity.（2）Software design of the microwave device.The software design includes the upper computer program design and the lower computer program design.The upper computer program and the lower computer program software are designed separately.The lower computer program is designed with STM32 microcontroller as the core to control the detection device to detect wheat in the frequency range of 2.5 GHz to 11.5 GHz and send the collected data to the upper computer;The upper computer performs data storage,curve plotting,data display and quantitative grain quality testing of the collected data.（3）Experimental study of AFB₁ detection in wheat based on microwave device.The transmission index of wheat samples with different AFB₁contents in the scanning frequency range were collected by a self-made microwave detection device and preprocessed by least squares filtering.A bootstrapping soft shrinkage（BOSS）algorithm was introduced for feature screening,and linear models（Partial Least Squares,PLS）and nonlinear models（i.e.,support vector machine,SVM;extreme learning machine,ELM;random forest,RF）for rapid and quantitative determination of AFB1 content in wheat,and the model performance was compared using randomization tests.The research results show that the BOSS feature optimization algorithm can select highly targeted feature variables,reduce the number of features input to the model,and effectively reduce the complexity of the model.Among them,the nonlinear model achieved better generalization performance compared with the linear model,and the BOSS-SVM model had the best performance among all models,with the root mean square error of prediction（RMSEP）,coefficient of determination（）and relative prediction deviation（RPD）in the prediction set of 2.8μg·kg^-1,0.97 and 5.7,respectively.The results show that the quantitative determination of AFB₁ content in wheat can be achieved with high accuracy using a self-made miniaturized microwave detection device combined with an appropriate chemometric method.In addition,the performance of the model can be further corrected to avoid visual errors by using a reasonable inspection technique.（4）Experimental study on the detection of moisture content of wheat based on microwave device.The transmission index of wheat samples in the frequency range of 2.5 GHz-11.5 GHz were acquired by the microwave device,and the transmission index pre-processed by least-square filtering were coarsely selected by using a competitive adaptive reweighted sampling（CARS）,and then the coarsely selected feature variables were finely selected by using a genetic algorithm（GA）for secondary features and ranked according to their weights.Support vector regression（SVR）was used to establish quantitative models with different combinations of feature variables for rapid detection of wheat moisture content.The results showed that the CARS-GA-SVR model had better detection performance for wheat moisture content determination than the CARS-SVR model.Among them,the CARS-GA-SVR model has the best detection when the combination of feature variables is 6-dimensional,with a RMSEP of0.4047%,aof 0.9756 and a RPD of 6.3234.