Development And Application Of Online Grain Moisture Content Detection Device Based On Multi-frequency Microwave Swept Measurement Technique

Grain is the material guarantee for human survival,and moisture is an extremely important chemical composition and processing indicator of grain.In the process of grain processing,the lack of accurate moisture content parameters will cause serious processing losses and mildew pollution.Compared with the traditional detection technology,the microwave technology has strong penetrability and can detect the accurate moisture content deep inside the grain layer,which is very suitable for the online detection of grain moisture content.However,problems such as low accuracy in high moisture content range,blind and subjective selection of detection frequency,and thickness fluctuations interfere with online detection limit the application of microwave technology,and solutions are urgently needed.In this study,on the basis of proving the response law of microwave characteristic parameters to grain moisture content,the ability of multi-frequency microwave to expand the moisture content detection range was discussed,and a multi-frequency microwave on-line detection device for grain moisture content was designed and developed.Moisture content detection in the range of 15.45%～81.19%;on the basis of proving the correlation between microwave characteristics and grain moisture content in different frequency bands,a two-stage frequency selection framework is designed,which combines the heuristic algorithm primary selection and the integrated voting method to refine.The optimal microwave frequency for on-line detection is optimized by the framework.On the basis of clarifying the influence of different layer thicknesses on the microwave phase shift characteristics,a phase shift correction algorithm is proposed,which eliminates the influence of thickness fluctuation interference on detection accuracy,and realizes the Thickness-independent on-line detection of grain moisture content.The main research contents and conclusions are as follows:(1)A microwave grain moisture content detection device was developed.The propagation law of microwave in grain lossy medium was studied,and the response law of microwave characteristic parameters to grain moisture content was explored,and then a static detection test platform and an online detection device were designed and developed.The two sets of devices are composed of microwave antennas,signal transceiver and demodulation circuits,small microwave anechoic chambers,and data acquisition systems.The mechanical structure of the two sets of devices was designed by Solidworks,and the data acquisition program was written by Keil.The static detection test platform was developed in the early stage,and the detection frequency is 2.60～3.00 GHz in the S-band,which is used to verify the detection principle and explore the optimization method.In the later stage,the online detection system was developed,and the detection frequency was expanded to 2.00～10.00 GHz in the S to X frequency bands,and it was equipped with a laser thickness sensor IX-150 and a special industrial computer to realize the online detection of grain moisture content.The online detection system application software Moister is developed based on the Qt platform,and the pre-trained model is called,which can be applied to the moisture content detection of different grains.(2)Research on static detection of grains with high moisture content based on multi-frequency microwave scanning measurement technology.Using the microwave grain moisture content static detection test platform,multi-frequency microwave scanning measurement was performed on fresh corn with high moisture content,and the microwave attenuation and phase shift characteristic data at different frequencies were collected;using the deep neural network(DNN)algorithm,the different The correlation between microwave characteristics and corn moisture content under frequency was compared,and the advantages and disadvantages of single-frequency microwave measurement method and multi-frequency microwave scanning measurement method were compared.The results show that there are significant differences in the correlation between microwave characteristics and corn moisture content at different frequencies,and the R~2 of the model with inferior frequencies is only 0.613;the multi-frequency microwave scanning measurement method is better than the conventional single-frequency microwave measurement method,and the index of the multi-frequency model is better.For any single frequency model,acceptable indicators(R~2=0.930,RMSE=1.552%,MAE=1.247%)are obtained in the high moisture content range;proper frequency selection can help improve model performance.(3)Research on on-line detection of grain moisture content based on second-order frequency selection framework.The microwave characteristic spectrum of corn with different moisture content was collected by the online detection device,the spectral range and standard deviation indexes were analyzed,and the correlation between microwave characteristics and grain moisture content in different frequency bands was explored.Furthermore,a new two-stage frequency selection framework(TSFSF)is designed to optimize the optimal microwave frequency for online detection.The moisture content detection model was constructed by using multiple linear regression(MLR),support vector machine(SVM),random forest(RF),adaptive boosting(Ada Boost),extreme gradient boosting(XGBoost)and DNN algorithms,and proposed Friedman-Nemenyi Joint test method to determine the best model for online detection.The results showed that the correlation between microwave characteristics and corn moisture content in different frequency bands was quite different,and the standard deviation SD fluctuated significantly.In the generation stage of TSFSF,17 candidate frequency subsets are generated using random forest-recursive feature elimination algorithm(RF-RFE),which are distributed in three dominant frequency bands of 2.00～4.00 GHz,6.00～8.00 GHz,8.00～9.00 GHz;In the decision-making stage of TSFSF,the cross validation-majority voting method(CV-MVM)is used to select the best frequency set,including 8 frequencies of 2.86,2.92,2.97,3.00,3.04,3.32,3.70 and 6.54 GHz.The Friedman-Nemenyi joint test selects the DNN model as the best model and deploys it to the online detection device.The MAE of the repeated test results does not exceed 1.50%and the SD does not exceed 1.76%,showing good accuracy and repeatability.(4)On-line detection of moisture content independent of thickness.The microwave characteristic spectrum of paddy with different moisture content under different stacking thickness was collected by online detection system.Based on the microwave phase-shift spectrum results,the law of phase-shift measurement is analyzed,and a phase-shift correction algorithm PSCA is proposed to solve the phase-shift ambiguity problem.Based on TSFSF and frequency set scale coordination mechanism,the optimal frequency for online detection of rice moisture content was explored.A high-precision sensor was introduced to collect the data of paddy layer thickness to establish a thickness compensation model to deal with the misleading attenuation caused by thickness fluctuations.The results show that the microwave phase shift feature has a negative value,and with the increase of the measurement frequency,the absolute value of the phase shift value increases,and the phase shift spectrum shows a monotonous downward trend.At the same time,when the sample layer thickness and moisture content increase,the will lead to an increase in the equivalent electrical length of the sample,resulting in a larger phase shift.The phase shift correction algorithm PSCA can effectively solve the phase shift ambiguity problem and obtain the actual phase shift value generated after the microwave penetrates the sample.TSFSF obtained the dominant frequency bands for on-line detection of rice moisture content,which are 2.00～3.00GHz,7.00～8.00 GHz,8.00～9.00 GHz,and the trigger frequency set scale coordination mechanism(FSSCM)searched for the best alternative subsets,including 2.25,2.26,2.27,2.28,2.29,2.30,2.31 and 8.94 GHz.The phase shift correction algorithm combines the thickness compensation modeling method to obtain a moisture content detection model that is independent of thickness.The SVM model selected by the Friedman-Nemenyi joint test faces the fluctuation of paddy thickness,and the detection result is always near the reference value,and the maximum SD value is only There is 0.537%,which realizes the on-line detection of moisture content independent of thickness.