Real time detection of soil moisture in winter jujube. orchard. based on NIR spectroscopy

Transcription

1 Real tme detecton of sol mosture n wnter jujube orchard based on NIR spectroscopy Xaofe An 1, Mnzan L 1 *, Lhua Zheng 1, Yumeng Lu 1, Yajng Zhang 1 Key Laboratory of Modern Precson Agrculture System Integraton Research, Chna Agrcultural Unversty, Mnstry of Educaton, Bejng , Chna Edtoral offce of Transacton of CSAM, CAAMS, Bejng, Chna *Emal: pact@cau.edu.cn Abstract. The measurement and control of sol mosture are the key technologes of precson agrculture. In order to real-tme detect sol mosture content faster and more accurately, a portable sol mosture sensor based on NIR spectroscopy was developed. Wth the sxty sol samples collected from a wnter jujube orchard, a lnear regresson model was establshed. The determnaton coeffcents of the calbraton ( R c ) and valdaton ( R v ) reached 0.88 and 0.9, respectvely. The model passed F-test and t-test and showed robust. Subsequently, two spatal dstrbuton maps of sol mosture were generated based on the data obtaned by the portable sol mosture detector and the data obtaned by oven dryng method, respectvely. Fnally, the correlaton between these two maps was nvestgated by usng the software of Surfer 8.0. The zones of dry and wet sol could be dstngushed easly n both maps. The results of the study showed that the developed detector was practcal. Keywords: spectroscopy Sol mosture, Real tme detecton, Sol sensor, Near nfrared reflectance

2 1 Introducton Sol mosture content s one of the most crtcal sol components to plant growth and land management, especally n dryland [1]. It s usually measured by the oven dryng method wth a qute hgh precson. On the other hand, the oven dryng method s a dffcult, costly and tme consumng procedure, so that t s necessary to develop a real tme sol mosture detector for the precson agrculture practce. The near-nfrared (NIR) reflectance spectroscopy s a promsng non-contact measurement method avalable to provde rapd nformaton about some sol physcal and chemcal propertes, such as sol organc matter content, sol total ntrogen, potassum, phosphorous, ph, sol mosture and partcle sze et al. When understandng the nteracton between the NIR ncdent lght and the sol propertes, those physcal and chemcal parameters can be estmated based on Chemometrcs. Many scholars such as Shbusawa, Chrsty, Xao W [-19] have researched on estmaton of sol mosture based on NIR spectroscopy n recent years. Mouazen [] establshed a sol mosture forecastng model wth correlaton coeffcent of 0.98 n the laboratory. The model method was partal least squares regresson (PLSR) wth 306.5~ nm. But n the feld measurement, the correlaton coeffcent was just 0.75 and the maxmum error of measurement was above.5%. Hummel [3] analyzed relatonshp between sol mosture and near nfrared spectroscopy, and a model was establshed by the multple lnear regresson (MLR) method. Peng Y.K. [4] adopted the NIR spectra (Neotec51A) to forecast sol mosture, organc matter and total ntrogen of Loess Plateau and also obtaned several estmatons models. Shbusawa et al. reported a development of a real-tme NIR spectroscopy measurement system for sol mosture [8]. The lght llumnaton and reflecton fbre were attached to a spectrophotometer wth nm VIS and NIR lght source. Most of above researches establshed forecastng models n the laboratory condtons, and the real-tme detecton of sol mosture has not been realzed wth hgher practcal applcablty. Thus, we decded to develop a portable and relable sol mosture sensor for the real-tme management. A comparson between thematc maps of sol mosture obtaned by the developed detector and oven dryng method respectvely was conducted. And the feld experments were executed n a wnter jujube orchard to test the performance of the detector.

3 Materals and methods The wnter jujube orchard was located at Bejng suburb, wth the area of 1000 m. Cultvated varety was Zhanhua wnter jujube planted as south-north row drecton. The range of experment zone was E ~ and N ~ The sol texture was typcal loam n north Chna. The detector was desgned wth an optcal unt and a control unt. Fg.1 shows the overall structure of the sol mosture detector, and Fg. shows the prototype of the detector. The optcal unt ncluded a near nfrared lamp-house at 1450 nm, a shared lamp-house drve crcut, a Y-type optcal fber shared by ncdence and reflectance, a probe and an InGaAs photoelectrc sensor. The control unt ncluded an amplfer crcut, a flter crcut, an A/D crcut, a LCD dsplay crcut and a storage crcut wth U-dsk. Lght from near nfrared lamp-house was adopted n the detector nstead of the sunlght. When the detector worked, t was pushed nto sol under the surface of 30 cm frstly. Then the optcal sgnal at sngle wavelength was transferred from the 1450 nm near-nfrared lamp-house to the surface of the target sol va the ncdence optcal fber. The reflected lght from the sol surface was acqured and transferred to the InGaAs photoelectrc sensor, where the optcal sgnal was converted to the electrcal sgnal. Subsequently, the obtaned electrcal sgnal was processed and stored temporarly by an 89S5 MCU. Fnally, the calculated sol mosture content was dsplayed on a LCD and stored n a U-dsk at the same tme. 1450nm lamp house Communcaton Incdent lght Target sol Data processng Reflected lght Photoelectrc sensor Electrcal sgnal Collectng data Optcal unt Fg.1. Overall structure Control unt

4 Handl LCD Power swtch Manual storage swtch Communcaton Optcal fber Flag on 30 cm Probe Fg.. Prototype of the detector Fg.3 shows the sgnal processng flow n the control unt. After the I-U convertng crcut, the optcal sgnal I 0 was converted to the electrcal sgnal U 0 by the photoelectrc sensor. The electrcal sgnal U 0 was a weak sgnal. It was necessary to desgn an amplfer crcut and a flter crcut before A/D crcut. The processed sgnal U 3 was transmtted to the MCU 89 S5. In the MCU 89S5, the data were dsplayed wth LCD and at the same tme stored n the U dsk. And fnally the data could also be transmtted to the PC by a seral port.

5 Lght I-U crcut sgnal I 0 U 0 Amplfer crcut Dsplay wth LCD U 1 Flter crcut U A/D crcut U 3 MCU 89S5 Seral comm uncaton Storage wth U dsk Fg.3. Sgnal processng flow In ths detector several modules were ncluded n the software, man program, subroutne of A/D converson, subroutne of data processng, subroutne of LCD dsplay and subroutne of storage. Fg.4 shows the flow chart of the man program.

6 Fg.4. Flow chart of man program Absorbance was selected as the spectral parameter. In order to get the absorbance of sol samples, a standard whteboard was desgned. Frst, the probe of the developed sol mosture detector was put on the standard whteboard and the output voltage V of the nstrument was measured. And then the probe was pushed nto sol sample and the output voltage ' V of the nstrument was measured. Accordng to the Equaton (1), () and (3), the reflectance absorbance calculated. A and average absorbance ' r, A avg of every sol sample were r V / V 100% (1) A lg( 1/ r ) () 6 A 6 (3) avg A n 1 3 Results and dscusson

7 3.1 Data preprocessng Sol mosture contents of all sol samples were also measured by the oven dryng method n the laboratory condton. Sol samples were subjected to oven dryng at 60 C for 4 hours, and sol mosture content was obtaned through the Equaton (4). m1 m0 w 100% (4) m 0 where: w s sol mosture content, m1 s orgnal sol weght, m0 s sol weght after oven dryng. Statstc characterstcs of sol mosture contents n the wnter jujube orchard are shown n Table 1. All sol samples were dvded nto two groups: calbraton group (ncludng 40 sol samples) and valdaton group (ncludng 0 sol samples). In the calbraton group, the selected sol mosture range of 3.15%~14.19% covered the most acqured plant and organc actvtes durng the wnter jujube season. In the valdaton group, the range was 4.54%~1.09%, and the average was 8.03% and the standard devaton was.09%. Table 1. Statstcs of sol samples Parameter Calbraton group(40) Range Average Standard devaton Valdaton group(0) Range Average Standard devaton Sol mosture (%) 3.15~ ~ Estmaton modelng of sol mosture The object of ths task was to buld a statstcal model between the absorbance of 1450nm and sol mosture content. There were several statstcal modelng technques for the proper calbraton performance, such as the lnear and non-lnear multple regresson analyses and the prncpal component analyss (PCA). In ths work, the forecastng model of sol mosture content was establshed

8 by the lnear regresson method wth the average absorbance spectral data. Fg.5 shows the comparson between measured data and predcted data of sol mosture. The forecastng model s shown n Equaton (5). The determnaton coeffcent of the calbraton ( R ) was 0.88, and the determnaton coeffcent of the valdaton c R ) was 0.9. The model passed the F-test and t-test, whch llustrated that both ( v the model and the parameters were sgnfcant. y x (5) where: y represents the dependent varable of sol mosture content; x 1450 s average absorbance of sol at the wavelength of 1450 nm. N N SM Met 0 1 SM error Met 0 1 (a) Predcted SM dstrbuton (b) Measured SM dstrbuton (c) SM error dstrbuton Fg.6. Thematc maps of sol mosture 3.4 Dscusson Use of nfrared emttng dode of 1450 nm could make t possble to ncrease the frequency of samplng at a faster rate. It would also make the unt run on low power, compact and lght, so that t would requre lttle space and mechancal support. Sol mosture content could be obtaned by the sol mosture sensor. A lnear regresson model could satsfy the basc needs. The determnaton coeffcent of the calbraton ( R c ) and valdaton ( R v ) were 0.88 and 0.9 respectvely. Root mean square error of calbraton (RMSEC), root mean square error of predcton (RMSEP) and resdual predcton devaton (RPD) were 0.94, 0.85 and 3.6 respectvely. From the thematc maps of sol mosture, the zones of dry and wet sol could be dstngushed easly. Both the results ndcated that the sensor could

9 meet the requrement of precson agrculture. In order to ncrease the accuracy of the NIR measurement, other nfluencng parameters such as sol texture and densty should be taken nto account. Both the texture and densty affected the results serously. Extendng the sensor applcaton for all sol textures requres ncorporatng extra laboratory measurements of spectra taken from sol samples of another sol texture. 4 Conclusons Sol mosture content s one of the most crtcal sol components to plant growth and land management, especally n dryland. In ths paper, a rapd method to measure sol mosture was proposed. Accordng to the result and dscusson, the followng conclusons could be obtaned. (1) A portable sol mosture detector was developed based on NIR spectroscopy. In the detector, LED of 1450 nm, Y-type fber and crcuts were adopted to nsure the sensor stable and relable enough. It also made the unt run on low power, compact and lght so that t would requre lttle space and mechancal support. () The feld experment showed that the nstrument could work stably. The establshed model showed that there were hgh correlaton coeffcent between average absorbance of 1450 nm and sol mosture content, wth R c and R v of 0.88, 0.9 respectvely. Smlar spatal patterns of sol mosture measured wth the developed detector and oven dryng method were observed. Both above results ndcated that the detector could satsfy the basc agrculture requrements. (3) In order to mprove the accuracy of sol mosture detecton, sol texture should be pad more attenton n the next research step. Acknowledgements Ths study was supported by Chnese Natonal Scence and Technology Support Program (011BAD1B01), NSFC projects ( , and ), and 863 programs (011AA100703, 011AA100704). References 1. Shang, S.P., Wang, Y.L., Hong A.J. Establshment of calbraton equaton for

10 sol mosture content by NIRS. Chnese Journal of spectroscopy laboratory. 8(4): (011).. Mouazen, A.M., Baerdemaeker, J.D., Ramonb, H. Towards development of on-lne sol mosture content sensor usng a fber-type NIR spectroscopy. Sol & tllage research. 80(1), (005). 3. Hummel, J.W., Sudduth, K.A., Hollnger, S.E. Sol mosture and organc matter predcton of surface and subsurface sols usng an NIR sol sensor. Computers and electroncs n agrculture.3(), (001). 4. Peng, Y.K., Zhang, J.X., He, X.S. Analyss of sol mosture organc matter and total ntrogen content n loess n chna wth near nfrared spectroscopy. Acta Pedologca Snca. 35(4), (1998). 5. Chrsty, C.D. Real-tme measurement of sol attrbutes usng on-the-go near nfrared reflectance spectroscopy. Computer and electroncs n agrculture. 61,10-19(008). 6. Gashaw, A., Shane, M.W. Development of a prototype nfrared reflectance mosture meter for mlled peat. Computer and electroncs n agrculture. 8,1-14(000). 7. Xao, W., L, X.Y., L, P.W. Adaptablty of the model for sol mosture measurement based on near-nfrared spectroscopy. Transactons of the Chnese Socety of Agrcultural Engneerng, 5(3), 33-36(009). 8. Shbusawa, S., Made, I., Anom, S.W. Ste-specfc crop response to temporal trend of sol varablty determned by the real-tme sol spectrophotometer. Proceedngs of the Jont European Conference of ECPA-ECPLE. Wagenngen Academc Publshers, Berln, Germany, pp (003). 9. Lang, X.Y., L, X.Y., Le, T.W. Study of sample temperature compensaton n the measurement of sol mosture content. Measurement. 44(10),00-04(011). 10. Lang, X.Y., L, X.Y., Wang, W. NIR Spectroscopy Based on DWT and LS-SVM for Predcton of Sol Mosture. Proceedngs of the 010 Internatonal Conference on Electrcal and Control Engneerng (ICECE 010), (010). 11. Mnasny, R., Alex, B.M., Veronque, B.M. Removng the effect of sol mosture from NIR dffuse reflectance spectra for the predcton of sol organc carbon. Geoderma, 167,118-14(011). 1. Pan, L. Development of sol organc matter fast determnaton nstrument

11 based on spectroscopy. Chna Agrcultural Unversty, L, M.Z., Pan, L., Zheng, L.H., An, X.F. Development of a portable SOM detector based on NIR dffuse reflecton. Spectroscopy and Spectral Analyss.30(4), (010). 14. L, M.Z., Han, D.H., Wang, X. Spectral analyss and applcaton. Scence Press, Bejng(006). 15. Chen, H.H., L, X, Dng, W.X. Twelve Knds of Grddng Methods of Surfer 8.0 n Isolne Drawng. Chnese Journal of Engneerng Geophyscs. 4(1),5-57(007). 16. L, X.Y., Xao, W., L, P.W. Near-nfrared spectral detecton of sol mosture based on feature extracton of FFT. Transacton of the Chnese Socety for Agrcultural Machnery.40(5):64-67(009). 17. Lang, X.Y., L, X.Y. The Research on Sample Temperature Compensaton for Sol Mosture Content Based on Mult-Wavelength and LS-SVM. Chnese Journal of Sensors and Actuators.4(8),18-13(011). 18. Krapez, Jean-Claude, Oloso, A combnaton of temperature, vegetaton ndexes and albedo, as obtaned by arborne hyperspectral remote sensng, for the evaluaton of sol mosture. Qrt Journal.8(),187-00(011). 19. Song, T., Bao, Y.D., He, Y. Research on the Method for Rapd Detecton of Sol Mosture Content Usng Spectral Data. Spectroscopy and Spectral Analyss. 9(3), (009).