AU1757801A

AU1757801A - Optical analysis of grain stream

Info

Publication number: AU1757801A
Application number: AU17578/01A
Authority: AU
Inventors: Suranjan Panigrahi; Guangjun Zhang
Original assignee: North Dakota State University Research Foundation
Current assignee: North Dakota State University Research Foundation
Priority date: 1999-11-08
Filing date: 2000-11-08
Publication date: 2001-06-06
Anticipated expiration: 2020-11-08
Also published as: WO2001035076B1; EP1147395A1; WO2001035076A1; AU774890B2

Description

WO 01/35076 PCT/USOO/30627 OPTICAL ANALYSIS OF GRAIN STREAM This application claims priority of U.S. provisional patent application No. 60/164,161 filed November 8, 1999. 5 FIELD OF THE INVENTION The present invention relates to a method and apparatus for optically analyzing a stream of an agricultural product in order to determine constituents of the product. BACKGROUND OF THE INVENTION 10 Systems are known in the art for the optical analysis of a stream of grain. As the grain is harvested in the field, a light source passes light through the grain stream. The transmitted light is detected by a receiver and processed by a computer under software control. By comparing the spectral absorption with values representing known absorption, the grain can be analyzed to determine its constituents. A need exists for improvements in 15 the optical analysis of agricultural products. SUMMARY OF THE INVENTION An apparatus consistent with the present invention measures constituents of an agricultural product. In the apparatus, a device forms a stream of the agricultural product. An optical sensing window in the device passes the stream of agricultural product, and a 20 radiation source contained within the housing irradiates the stream of agricultural product as it passes through the optical sensing window. A receiver receives radiation transmitted through the stream of agricultural product and converts the received radiation into a corresponding electronic signal. A processor, coupled to the receiver, receives the electronic signal and analyzes it under software control to determine the constituents of the agricultural 25 product. BRIEF DESCRIPTION OF THE DRAWINGS FIG. I is a block diagram of a system for optically analyzing a stream of an agricultural product. FIG. 2 is a diagram of a side cover in the system of FIG. 1. 30 FIG. 3 is a perspective diagram of a detector box in the system of FIG. 1. FIG. 4 is a perspective view of a sensing window in the system of FIG. 1. FIG. 5 is a top view of the sensing window. FIG. 6 is a side view of the sensing window.

WO 01/35076 PCT/USOO/30627 2 FIG. 7 is a front view of the sensing window. DETAILED DESCRIPTION Optical Analysis System 5 A system to predict the protein/constituents of agricultural products is described below with reference to FIGS. 1-7. As shown in FIG. 1. an inlet I of the material handling system includes specially designed pipes and accessories. It can be attached to an auger, a clean grain elevator or any outlet of a storage bin. The grain or product entering through inlet 1 moves through grain passage 3. bounded by inner transparent wall 4 and metallic outer wall 10 5. The grain entering through inlet I passes through a sensing window 6, which has a definite thickness. A position switch 2 is connected to a control unit 7, which is also connected to an electric motor 8. operated by Direct Current (DC) power source 9. Electric motor 8 is mounted within an enclosure box of an auger 10 containing discharger auger 11. Auger 11 is driven by motor 8. Auger I 1 through an outlet 12 can discharge the grain out 15 from the system back to the original stream of grain or any user-defined location. An illumination chamber 13 is bounded by transparent wall 4 and vertical opaque wall 17. A base 14 is mounted on a wall 17. A lamp (illumination source) 16 is attached by a lamp holder 15 and is connected to the power source and control box 19 through a cable 17 A. Sensor body 21 includes air inlet passages 43. Lamp 16 may be implemented with, for 20 example, a tungsten-halogen lamp. Sensor body 21 is attached with a DC fan 20. A side cover 22 of sensor body 21 (shown in FIG. 2) has a small fan 23 mounted on the cover and is operated by DC power source 24. A sensor head 32 is composed of optical passage 25, optically isolated from outer environment by metallic covers 33 and a detector box 26. Sensor head 32 is attached to 25 sensor body 21 by a mount 3 1. The tip of fiber optic probe 27 is mounted on detector wall 34 (FIG. 3) of detector box 26. Detector box 26 is composed of a front lens wall 35. a detector wall 34, a base plate 37. and a top cover 36 (FIG. 3). Front lens wall 35 contains two lenses. 41 and 42 (in series) arranged between three retainers 38. 39. and 40. Fiber optic cable 28 is connected to a portable spectrometer 29 including a diffraction 30 grating and an array of charged coupled device (CCD) detectors. Spectrometer 29 is coupled to with a computer 30. Operation The grain or agriculture product enters through inlet 1, and it fills up grain passage 3- WO 01/35076 PCT/USOO/30627 3 A and 3 and the empty space in the auger. When the level of grain reaches the level at which position or proximity sensor 2 is located, position sensor 2 triggers the auger to run auger 11. The running of the auger allows the grain to move through auger 11 and out from the system through outlet 12. The location of position sensor 2 along with features of wall 4, sensing 5 window 6, grain passage 3 and 3-A, and auger 11 allow the grain to move at a constant rate. This feature also helps to eliminate dust build-up on the inner wall of sensing window 6. Dust build-up can adversely affect performance of the sensor. The near infrared (NIR) beam contained in the emitted illumination by light 16 transmits through the flowing grain in the sensing window 6. The transmitted light passes through optical passage 25 and 10 subsequently passes through detector box 26. The front wall of optical system 35 of detector box 26 converges the transmitted light or radiation to fall on the tip of fiber optic probe 27. The transmitted light/radiation is conveyed through fiber optic cable 28 to portable spectrometer 29. Spectrometer 29 with the use of the computer 30, under software control, records the spectral signature of the transmitted light or radiation between 700-1100 15 nanometers (nm). Sensing Window FIG. 4 is a perspective view providing more detail of sensing window 6 in the system of FIG. 1. FIGS. 5-7 are, respectively, top, side, and front views also providing more detail of sensing window 6. As shown in FIGS. 4-7, sensing window 6 is formed by inner 20 transparent wall 4. and outer sensing wall 6-B, located behind metallic outer wall 5. The two side walls 6-B of the sensing windows are composed of opaque materials, and they connect the inner and outer walls. Grain passage 3 is empty space formed by the inner transparent wall 4, outer sensing wall 6-B. and two side walls 6-A. A circular area 6-C on the metallic outer wall 5 defines the effective sensing region through which the transmitted beam passes 25 to the sensing head. Software Processing Computer 30 may use a number of software-implemented techniques and algorithms to process the signal output by spectrometer 29 and determine constituents of the agricultural product. Examples of those techniques and algorithms are explained in Appendix A. 30 While the present invention has been described in connection with an exemplary embodiment, it will be understood that many modifications will be readily apparent to those skilled in the art, and this application is intended to cover any adaptations or variations thereof. For example, different types of materials for the device, and various types of WO 01/35076 PCT/USOO/30627 4 software algorithms for processing the signal resulting from irradiation of the agricultural product, may be used without departing from the scope of the invention. This invention should be limited only by the claims and equivalents thereof. 5 10 15 20 25 30 WO 01/35076 PCT/USO0/30627 5 The advaitsges of these algorithms: At present, the conventional technique in measuring/predicting the concentration of desired constituent (i.e. protein., or oil content) using NIR technique involves the transmission/ reflectance of a reference sample. To translate this process for on-the-go sensor could create problem. It will be difficult to use a separate reference sample (other than the product) to obtain reference signal to be used for predicting the constituent's contribution in real-time/on-the-go basis. Thus we have proposed algorithms/techniques that could eHminate the need for having a separate reference sample (other than the product) and thus could be used with on-the-go sensor, and could make the design, development and operational process of the sensor more efficient. (Continued.....) VYTT'ITTTTY7 IYYTTVI' /YYTr w - WO 01/35076 PCT/USOO/30627 6 Algorithm Nomenclature: Raw transmission or (signal-dark) reflection signal 0 0 wavelength Figure I

X

1 . X 2 , X. ......... - wavelengths critical for predicting the constituent of product. or wavelengths with highest or higher correlation with the concentration of constituent. -ro- reference wavelength, that does not contribute to the concentration of the desired constituent or does not have any correlations with the concentration of the desired constituent Algorithm 1: Xt, X2. A3........... ., or A could be obtained from prior experiments, literature or be determined for a given equipment and/or for a given agricultural product with specific to the desired constituent. They can also be determined by conducting experiments and using statistical, or other data grinning techniques such as neural network/genetic algorithms. (Continued.....) SUBSTITUTE SHEET (RULE 26) WO 01/35076 PCT/USOO/30627 7 1. Obtain the raw spectra of a sample 2. Subtract the dark signal from the raw spectra to obtain [SuA ] (spectra - dark) , . ................. (1) 3. Nonnalize the (spectra - dark) signal, Sux, by Ad using the following relationship. [ . ul [su L (2) where, Sna = normalized signal from X = k to X - L As - nonnalizing wavelength SAd - spectral signal at normalizing wavelength As X can bc )L, or X. or X 3 . . . . . . . , or , S5 can also be P where, P -f(7 1 , X2, 3......... I) 4. Normalized spectra, Sna, or its first or second or any other higher order derivative along with suitable statistical or neural network based prediction technique can be used to predict the concentratiou of the desired constituent Corollary I : - At step 3. in addition to normalizing, Sut, additional linear or non-linear processing of spectra, SnA. could be possible. a At step 3, before norrnalizing, the (spectra-dark) signal, Sul, can be processed in any linear/non-linear way. a At step 3. in equation (2). Sij can be replaced by SaU S'a = S,, for X = q .. ,t; S2 - Signal at a given waveleogth, A .......... (3) or SAb = P. where, P = f(......... (Continued....) CTTDCTTTTT' QUFVUT I/TDI r WO 01/35076 PCT/IUSOO/30627 8 Signal S, ----- q t ...... o A, wavelength Band of wavclcngth centered around X Figure 2 (Continued.....) SUBSTITUTE SHEET (RULE 261 WO 01/35076 PCTUSOO/30627 9 Flow Chart (speca-al - dark) signal, SUL. - 7 ---- 14 X. Protein content or other constituent (Continued... SUBSTITUTE SHEET RULEE 261i WO 01/35076 PCT/USOO/30627 10 Corollary II: Another corollary (Corollary U1) is described below: Xi, 1 2 , Xs ------ '-- wavelengths critical for predicting the constituent of product (figure 1) la2 - transmitted radiation at X2 1a - transmitted radiation at X1 IX2 - incident radiation at X2 lax, - incident radiation at Xl Thickness of sample, I, of the product, whosc constituent is being measured Sensing window Figure: 3 Algorithm 2: 1. Obtain dark signal of the set up (with a given light source, fiber optics, spectrometer). 2. Obtain L, and Ix,. These can be obtained without using any sample in the sensing window. 3. Obtain Ix, and Ix . (The sample needs to be there) in the sensing window. 4. Determine signal [d -ark1 S' = log, -dark I -dark-------(4) I. -dark j 5. Se can be used to predict the constituent of the given product, using a suitable neural network or statistical prediction model. (Continued.......) 1T TRqTITYTTF qHFFT (12TT, 7i6 WO 01/35076 PCT/USOO/30627 11 Algorithm 3: This algorithm could eliminate the need for taking separate reference signal for the on-the-go sensor. Sensor head Spectrometer .- F Fiber opties ----- Sensing window Tllumination Source Figur: 4 1. For the given setup of illumination. sensing window, sensor head, fiber optic and its associated setting, find dark signal. 2. Using the available gating mechanism (which can be automatically controlled), the intensity of light is reduced and with no sample of the product, in the sensing window, find the spectral response from wavelength k to L. Let that be denoted by reference signal, [RAt. 3. Under running condition, the sample of the product will move through the sensing window. The gating mechanism will be adjusted back for the light to operate in the desired intensity. Obtain the transmitted signal, [Tt. 4. Subtract dark signal from both Reference signal, [R,= and transmitted signal, [TAIC 5. Obtain the normalized signal, [Snjt [Sn,-' = IT4 ] -dark signal RA t - dark signal 6. Process the normalized signal, [SnAr for reducing dimensional ities by averaging. 7. Find the second derivative of the normalized signal and further use the second derivatives to predict the concentration of the constituent using suitable neural network or statistical model. .qTTRSqTITTTTE TF HFVT (RUT 6 .

WO 01/35076 PCT/US00/30627 12 Algorithm 1, Corollary 1, was tested on wheat data collected for both static and dynamic conditions. > The raw spectral signal was obtained using PC-1000 and ,-2000 tiberoptics based spectrometer, (Ocean Optics Company, FL). The original data ranged from 682.67 rn 1212.37 nm (1100 data points). This signal was already subtracted from dark signal > Kept the data points between 699.67 nm and 1050.3 nm, (including both ends) thus keeping 698 data points. (used Microsoft Excel) > Reduced (preprocess) the data a factor of 4 -* I using commercial software, GRAMS/32. (Galactic Industries Corporation NH4). > Choosing reference wavelength. ). as 830 nm, A= A-NS. (This was done by Microsoft Excel) > The normalized signal, Sna, was obtained by dividing the processed signal SA. > The second derivatives of Sn; was obtained using commercial software. GRAMS/32, (Galactic Industries Corporation, NH). > The second derivatives of "Sn a" were used as the input to the Back Propagation Neural Network. Protein content was used as the output of the neural network. Two data sets were created. One set wa- the "training set" and the other was the "testing set". The inputs of the training and test set were normalized (with a mean of ' and standard deviation of '1'). Commercial software Professional II/Plus, (Neural ware, Pittsburgh. PA) was used to develop neural network. A neural network model (with momentum of 0.6 and learning coefficient of 0.3) was developed as the prediction model. The neural network was trained on training data set and tested on test data set. The performance of the prediction model in predicting protein content was evaluated by comparing the predicted protein content vs. actual protein content of the wheat samples. A stand-alone C4" program was written to calculate the overall % of accuracy, average absolute error, minimum absolute error and maximum absolute error. Absolute error - Absolute|(Actual protein content - Predicted protein content) Maximum absolute error = maximum value among all the absolute errors for the samples in the data set Minimum absolute error - minimum value among all the absolute errors for the samples in the data set SUBSTITUTE SHEET (RULE 261 WO 01/35076 PCT/USOO/30627 13 For the static condition, the described algorithm showed an accuracy of 96.17% and 93.80% (for training and testing data set respectively) with an average absolute error of 0.54 point of protein content (for training) and 0.84 point of protein content (for testing). For dynamic condition, the above algorithm showed an accuracy of 97.88% (for training) and 95.98% (for testing). The average absolute error was found to be 0.31 point of protein content (for training) and 0.59 point of protein content (for test data). SUBSTITUTE SHEET (RULE 26)

Claims

1. An apparatus for measuring constituents of an agricultural product, comprising: a device for forming a stream of the agricultural product; 5 an optical sensing window in the device for passing the stream of agricultural product; a radiation source contained within the housing for irradiating the stream of agricultural product as the stream of agricultural product passes through the optical sensing window; a receiver for receiving radiation transmitted through the stream of agricultural 10 product and for converting the received radiation into a corresponding electronic signal; and a processor. coupled to the receiver, for receiving the electronic signal and for analyzing the electronic signal to determine the constituents of the agricultural product.

2. The apparatus of claim 1 wherein the receiver includes: 15 a fiber optic cable; a sensing head for receiving the radiation transmitted through the stream of agricultural product and for focusing the received radiation onto the fiber optic cable; and a spectrometer, coupled to the fiber optic cable, for converting the received radiation into a corresponding electronic signal. 20

3. The apparatus of claim 2 wherein the fiber optic cable includes a single fiber optic cable.

4. The apparatus of claim 2 wherein the spectrometer includes: 25 a diffraction grating for dividing the receiving radiation into spectral components; and an array of charge coupled devices oriented to receive the spectral components.

5. The apparatus of claim 2 wherein the sensing head includes: a fiber optic probe coupled to the fiber optic cable; and 30 a plurality of optical lenses positioned between the optical sensing window the fiber optic cable for focusing the received radiation on the fiber optic probe.

6. The apparatus of claim 1. further including a housing for containing the SUBSTITUTE SHEET (RULE 26) WO 01/35076 PCT/USOO/30627 15 device, the radiation source. the optical sensing window, and the receiver.

7. The apparatus of claim 6, further including a fan mounted within the housing. 5

8. The apparatus of claim 1, further including an inlet, coupled to the device, for attachment to a source providing the agricultural product.

9. The apparatus of claim 6, further including a mounting location for accessories.

10 10. An apparatus for measuring constituents of an agricultural product, comprising: a device for forming a stream of the agricultural product; a radiation source contained within the housing for irradiating the stream of agricultural product as the stream of agricultural product passes through the device; a fiber optic cable; 15 a sensing head for receiving the radiation transmitted through the stream of agricultural product and for focusing the received radiation onto the fiber optic cable; a spectrometer, coupled to the fiber optic cable, for converting the received radiation into a corresponding electronic signal; and a processor. coupled to the spectrometer, for receiving the electronic signal and for 20 analyzing the electronic signal to determine the constituents of the agricultural product.

11. The apparatus of claim 10 wherein the spectrometer includes: a diffraction grating for dividing the receiving radiation into spectral components; and an array of charge coupled devices oriented to receive the spectral components. 25

12. The apparatus of claim 10 wherein the sensing head includes: a fiber optic probe coupled to the fiber optic cable; and a plurality of optical lenses positioned between the device and the fiber optic cable for focusing the received radiation on the fiber optic probe. 30

13. The apparatus of claim 10, further including a tungsten-halogen lamp for providing the radiation transmitted through the stream of agricultural product. SUBSTITUTE SHEET (RULE 26)