PURE SPECIES OF GRASS DISCRIMINATION WITH THE HELP OF HYPERSPECTRAL IMAGING NIR Laura Monica DALE,,, Ioan ROTAR1, Anca BOGDAN1, Florin PACURAR1, Andre THEWIS2,
Download ReportTranscript PURE SPECIES OF GRASS DISCRIMINATION WITH THE HELP OF HYPERSPECTRAL IMAGING NIR Laura Monica DALE,,, Ioan ROTAR1, Anca BOGDAN1, Florin PACURAR1, Andre THEWIS2,
PURE SPECIES OF GRASS DISCRIMINATION WITH THE HELP OF HYPERSPECTRAL IMAGING NIR Laura Monica DALE,,, Ioan ROTAR1, Anca BOGDAN1, Florin PACURAR1, Andre THEWIS2, Juan FERNÁNDEZ PIERNA3, Nicaise KAYOKA MUKENDI3, Vincent BEATEN3 Department of Grassland and Forage Crops, University of Agricultural Science and Medicine Veterinary, Cluj Napoca, 3-5, Calea Manaştur, 400372, Cluj Napoca Animal Science Unit, Gembloux Agro-Bio Tech, University of Liège, 2, Passage des Déportés, 5030 Gembloux, Belgium Walloon Agricultural Research Centre, Valorisation of Agricultural Products Department, 24 Chaussée de Namur, 5030 Gembloux, Belgium „Use of NIR HSI to detect the pure samples of a mixture” Arnica montana L. n= 4 (No =16) * Trifolium repens L. n= 4 (No =16) * Hieracium aurantiacum L. n= 4 (No =16) * Festuca rubra L. Agrostis capillaris L. n= 4 (No =16) * n= 4 (No =16) * - construction of spectral classes ; - identification of the spectral classes for pure samples; - discrimination of pure samples of a mixture of several species. * represents the number of preleveted samples from the category of forages (n x 4 =16) 2 Were build 4 classes for each type of pure samples: - Arnica montana L (AM) - Trifolium repens L (TR) - Hieracium aurantiacum L (HA) - Festuca rubra L(FR) 1000 spectra contein all data base. 3 Classes AM TR HA FR AM 100 0 0 0 TR 0 100 0 0 HA 0 0 100 0 FR 0 0 0 100 4 Classes AM TR HA FR AM 98.40 0.30 1.20 1.000e-01 TR 0 99.90 1.000e-02 0 HA 0.80 0.40 98.80 0 FR 0 0 0 100 5 The external validation model was build from 5 classes (4 pure species wich are in calibration model + Agrostis capillaris(AC)). 6 Samples/Scores Plot of X,c & Xac, 1.5 Y Predicted 2 (Class 2) 1 0.5 0 Y Predicted 2 (Class 2) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 2 x-axis zero y-axis zero -0.5 -1 200 400 600 800 Scores on LV 1 (88.48%) 1000 1200 1400 7 Samples/Scores Plot of X,c & Xac, 400 300 Scores on LV 2 (10.71%) 200 100 0 -100 Scores on LV 2 (10.71%) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 4 x-axis zero y-axis zero -200 -300 -400 -0.4 -0.2 0 0.2 0.4 0.6 Y Predicted 4 (Class 4) 0.8 1 1.2 1.4 8 Samples/Scores Plot of X,c & Xam, 2 Y Predicted 1 (Class 1) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 1 x-axis zero y-axis zero 1.5 Y Predicted 1 (Class 1) 1 0.5 0 -0.5 -1 500 Class AM AM 97.62 1000 1500 TR 0.79 2000 2500 Sample HA 1.59 3000 3500 4000 4500 5000 FR 0 9 Samples/Scores Plot of X,c & Xam, 150 Scores on LV 3 (0.37%) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 1 x-axis zero y-axis zero 100 Scores on LV 3 (0.37%) 50 0 -50 -100 -150 -200 -1 -0.5 0 0.5 Y Predicted 1 (Class 1) 1 1.5 2 10 Samples/Scores Plot of X,c & Xfr, 1.4 Y Predicted 4 (Class 4) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 4 x-axis zero y-axis zero 1.2 Y Predicted 4 (Class 4) 1 0.8 0.6 0.4 0.2 0 -0.2 -0.4 Class FR 500 AM 0 1000 1500 TR 0 2000 2500 Sample 3000 HA 0.60 11 3500 4000 FR 99.40 4500 5000 Samples/Scores Plot of X,c & Xfr, 400 300 Scores on LV 2 (10.71%) 200 100 0 Scores on LV 2 (10.71%) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 4 x-axis zero y-axis zero -100 -200 -300 -400 -0.4 -0.2 0 0.2 0.4 0.6 Y Predicted 4 (Class 4) 12 0.8 1 1.2 1.4 Samples/Scores Plot of X,c & Xhr, 1.5 Y Predicted 3 (Class 3) 1 Y Predicted 3 (Class 3) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 3 x-axis zero y-axis zero 0.5 0 -0.5 -1 Class HA 500 AM 1.37 1000 1500 TR 0 2000 2500 Sample 3000 HA 98.63 13 3500 FR 0 4000 4500 5000 Samples/Scores Plot of X,c & Xhr, 150 Scores on LV 3 (0.37%) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 3 x-axis zero y-axis zero 100 Scores on LV 3 (0.37%) 50 0 -50 -100 -150 -200 -1 -0.5 0 0.5 1 1.5 Y Predicted 3 (Class 3) 14 Samples/Scores Plot of X,c & Xtr, 2 1.5 Y Predicted 2 (Class 2) 1 0.5 0 Y Predicted 2 (Class 2) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 2 x-axis zero y-axis zero -0.5 -1 Class TR 500 AM 2.07 1000 1500 TR 96.73 2000 2500 Sample HA 1.19 15 3000 3500 FR 0 4000 4500 5000 Samples/Scores Plot of X,c & Xtr, 80 Scores on LV 4 (0.12%) Class 0 Class 1 Class 2 Class 3 Class 4 Discrim Y 2 x-axis zero y-axis zero 60 Scores on LV 4 (0.12%) 40 20 0 -20 -40 -60 -80 -1 -0.5 0 0.5 Y Predicted 2 (Class 2) 16 1 1.5 2 Classes AM AM 97.62 TR 2.07 HA 1.37 FR 0 TR 0.79 96.73 0 0 HA 1.59 1.19 98.63 0.60 FR 0 0 0 99.40 17 1. The discrimination or the underline of the pure samples, of the pure species is made using the Imaging NIR instrument (Camera NIR) with a very good standard prediction error SEP. 2. Because we can identify species with the help of the Near Infrared Hyperspectral Imaging or NIR Camera, it had been tried to relate the crude protein content with the organic substance’s digestibility content from samples of which’s floristical composition is known. So the calibration model can be used in order to discriminate species from mixes of two or three species. 19