Development Of Grain Moisture Detection Instrument Based On Near Infrared Spectroscopy

Detecting the moisture content of grain quickly and accurately and learning the changing tendency on time are of great significance to study the effect of water content on the food quality and characteristics during the storing period,which are one of the most important bases to implement informatization and intelligentization of modern agriculture.In order detect moisture content of different kinds of grains,this research designed and developed a grains moisture content detecting instrument and corresponding host computer measuring software based on the moisture content detection sensor probe,which was also designed by ourselves.Experiments were implemented to test the performance of the instrument and the results indicated that it could detect the water content of grains basically.The main contents of this research are as follows:?1?The basic principles of the near-infrared spectrum moisture detection,BP neural network technology,TCP/IP Ethernet technology and embedded operating system technology of?C/OS-II were analyzed,which provided a theory fundament for developing instrument.?2?This study designed a sensor probe based on near-infrared spectrum,which had the advantage over indirect moisture measuring and increase the precision and efficiency of on-line moisture detection.This study analyzed the relationship between moisture content of grains and spectral reflectance over full spectrum?1100¹250nm?1350¹550nm and1450nm based on PLS and CARS models,the results show that the correlation coefficient of the modeling set and the prediction set of the two methods at1350¹550 nm is above 0.95,RMSEP is also close to 0.016;the correlation coefficient at the 1450nm band is close to 0.90,and the RMSEP is close to 0.02;it is finally determined that 1450nm is the absorption peak of grain to water absorption.On this basis,the characteristics wavelengths of 1450nm and 980nm were employed as the characteristic light source and reference light source of the instrument.Optical elements such as filter,InGaAs photodiode and lens constituted the complete light path structure.?3?This study designed the spectral signal acquisition circuit,which could implement the optical current conversion,I/V conversion and signal amplification.This study designed temperature and humidity acquisition circuit,which could collect the temperature and humidity of environment when measuring the grain moisture.This study designed AD conversion circuit,which could implement AD conversion for the analog voltage data collected by the near-infrared sensor.This study designed time-sharing control circuit for data acquisition,which could control the collection of spectral signal acquisition circuit and temperature and humidity acquisition circuit.This study designed PHY Ethernet to control the circuit,upload and display the collected data on remote PC in real time,which provided an available means for analyzing data.This study designed a constant current source drive circuit aiming at light sources of sensors,which could provide stable driving current and reduce the measurement error.This study designed the man-machine interface circuit of touch and display integration,which implement the display and touch and the detecting data.?4?This study designed the program based on the embedded C language for the above circuits including the main control program,sensor control acquisition program,grain moisture detection program,touch display control program and Ethernet control program design.The combined hardware circuit and software program were debugged and the results satisfied the requirements of actual measurement.?5?This study implemented calibration test to detect the moisture content of unhusked rice based on BPNN models.The parameters of the three-layer BPNN models included the weight w₁₂2 from the input layer to the hidden layer and the weight w₂₃3 from the hidden layer to the output layer,and the threshold value b₂ of the hidden layer and the threshold b₃ of the output layer.The R² of BPNN was 0.9695.This model was transplanted to the instrument to test the performance.The R² was 0.9279 and the accuracy was 94.4%,which was in high precision.