| Our country is currently in the period of transformation from traditional agriculture to modern fine agriculture.The 13th Five Year Plan makes it clear that agriculture is the foundation of building well-off society and modernization.And we should accelerate the construction of modern agriculture and improve the quality of agricultural production.As a result,promoting the integration of information technology and agricultural production management as well as resource environment,improving the level of intelligent and precision agriculture is the practical way to improve the quality and safety of agricultural products and promote agricultural sustainable development.It must be said,the Internet of Things(IoT)is more and more applied in many links,such as the monitors of agricutural resource and ecological environment,delicacy management of agricultural production,transportation of agricultural products.Near-infrared(NIR)spectroscopy is regarded as one of the fastest-growing analysis technology in recent decades,which has three main features:rapid-response,non-destructive and simultaneous multi-element analysis.Integrating the NIR spectroscopy with IoT,we can develop a new kinds of spectral sensing node.The new sensor node can be used to build Spectral Sensing Internet of Things(SSIoT)which will have sensory ability of object’s composition.The focal points of this study includes mainly:(1)NIR spectroscopy and application;(2)the incidence optical systems design for spectral sensing node.In the part of NIR,the chemometrics algorithms(characteristic wavelengths selecting methods,pattern recognition models and quantitative regression models)were studied by taking green tea as research subjects,which provide technical support for the design and application of sensor nodes.In the part of optical design,a short-wave-infrared multichannel integrated optical spectrum assembly(MI OS A)designed by Shanghai Institude of Technical Physics was taken as the main light dispersion device and detector.By designing the optical and mechanical system,the detection range of sensor node was up to 50 m,which can be used to build SSIoT or ultralow-altitude remote sensing carried by an unmanned aerial vehicles(UAVs).The main researchs are listed as follows:(1)Identification of Shandong green tea origins,production date and intrinsic composition by near-infrared spectroscopy to analyze the effects on model precision of different modeling method and signal-to-noise ratio of spectra.The above study was aimed to explore the feasibility of non-destructive identification of green tea quality by low-resolution spectrometer.All spectra in the study were collected by a near infrared fiber optic spectrometer directly without samples pretreating,and the spectral band of 1300 nm-2300 nm was selected for further analysis.Specific results are as follows:①Signal-to-noise ratio(SNR)has great effect on analytical precision and the accuracy shows a third order function decreasing trend with the SNR decreased.When the classification accuracy achieves 85%and 90%,the required SNR are 142:1 and 261:1,respectively.②The origins identification model of Shandong green tea were built by principal component analysis(PCA),partial least squares(PLS),back-propagation artificial neural network(BP-ANN)and support vector machine(SVM),and the accuracy of all models were up to 95%excep PCA.Among them,PLS got the better result with 100%recognition rate.In addition,characteristic wavelength-discriminant analysis(CW-DA)method was first put forward and the prediction rate of green tea origins was up to 98%.③NIR was first applied in the identification of green tea production date.In the exprement,100 Rizhao green tea samples were collect to build PLS regression models,and each sample was given a label according to its production date order in the year.The correlation coefficient(Rcal)and root mean square error of cross-validation(RMSECV)of calibration set were 0.952 and 15.18.The correlation coefficient(Rcal)root mean square error of prediction(RMSEP)and relative percent deviation(RPD)of prediction set were 0.943,19.965 and 3.07,respectively.④ The regression models of tea polyphenols(TP),catechins(C),epigallocatechin gallate(EGCG)and epicatechin gallate(ECG)in green tea were built by PLS.Before model building,the characteristic wavelengths were selected by moving window-partial least squares(MW-PLS)and the predictive correlation coefficient were 0.943,0.705,0.814 and 0.68,respectively.The overall results indicated that low-resolution spectrometer can be used in non-destructive identification of green tea components,and the correlation coefficient are positively associated with the component content.(2)Moving window-back propagation artificial neural network(MW-BPANN)and induced mutation genetic algorithm(IMGA)were first proposed to select characteristic wavelengths of green tea origins identification models.The identification accuracy of SVM classification model was raised to 98.33%from 91.67%after characteristic wavelength selection by MW-BPANN.By contrast,11 characteristic wavelengths(1472.76 nm,1499.835 nm,1513.342 nm,1540.295 nm,1785.308 nm,1856.495 nm,1977.446 nm,1996.298 nm,2033.792 nm,2120.156 nm,2287.833 nm)were selected from 156 wavelengths and the classification accuracy was up to 100%.The experimental results show that the two methods can effectively select the characteristic wavelengths without physical and chemical analysis and improve the prediction accuracy.The selected characteristic wavelengths can inform the design of dedicated spectral sensing nodes.(3)Based on graphical user interface(GUI)development environment,a universal NIR analysis software(ARCO-NIR)was developed by M language.Adopting modular design ideas,ARCO-NIR is consist of for modules(preprocessing module,characteristic wavelength selection module,qualitative identification modul and quantitative regression module)according to NIR analysis process.The software has the advantages of friendly software interface and easy to use,which promotes quick learning for non-professionals.(4)To make up for the poor data fineness and detection efficiencies of space remote sensing and field spectrometer,a new spectral sensing node was developed with a MIOSA.The node can be used to build SSIoT or ultralow-altitude remote sensing carried by an unmanned aerial vehicles(UAVs).① A transrnissive incidence optical system was designed for sensor node working in passive illumination,which using variable diaphragm and cylindrical lens for field contral and beam shaping.The field angle was 1.8°-14.8°,the optical system size was Φ25.4 ×112.37 mm,and the effective detection range was 50 m.② A dynamic data acquisition and processing system was developed for sensor node,which was consist of hardware circuit and spectral data acquisition software.FPGA and MSP430 microcontroller were used in hardware for the whole project design.nRF905 was applied for wireless data communications and the power was supplied by a 5 V mobile power.The total system power consumption was less than 300 mW,and the wireless transmission distance was up to 150 m.Spectral acquisition software was deveioped by GUI programming and serial communication technology of MATLAB,which enables real-time spectral data collection,processing and display.③ Mechanical structure design,machining,assembly and debugging of the sensor node was carried out,including the lens cone,mechanical shell and other mechanical parts.All mechanical parts were aluminum which was modified by frosting oxidized treatment to reduce the influence of stray light.Assembly and debugging was accomplish by using snap ring wrench and three-dimensional precision translation.A sensor node prototype was got with the dimension 122 mmx91 mm×58 mm.④ Calibration and performance testing was completed for the sensor node.Wavelength calibration and resolution was completed by a monochromator(Omni-A5008),and wavelength precision was tested by a standard glass(SRM2350a)and xenon lamp,respectively.The results show that the wavelength range was 1355 nm-1565 nm,the wavelength precision was 2 nm,and the resolution was 7 nm.Absorbance accuracy was validated by samarium oxide powder,which was coincided with the test result of Avantes spectrometer.The outdoor experiment(light intensity:34940 Lux,field:2.7°,integration time:20 ms)indicated that the sensor node can collected the leaves spectra 50 meters away.(5)To overcome the shortfalls of transmission incidence optical systems in miniaturization and lightweight design,a reflective optical system of sensor node was proposed which was composed of microlens array,reflection cup and cylindrical mirror.The microlens array was used to limite field and modulate the distribution of incident light.The reflection cup and cylindrical mirror was applied for beam shaping.In the paper,the calculation formula for calculating circular size of optical system was derived by infrared absorption model,detecting range,water quality and etc.And then,the focal length and calibre of primary mirror was calculated.Next,the design of optical system structure was completed by ZEMAX,which was further ray tracing simulated by TracePro with one million lights.The analysis results showed that the designed system satisfy the requirements in both spot size and energy.The overall dimension(Φ24.572 mm × 34.3 mm)meet the purpose of the micromation. |
PDF Full Text Request