Академический Документы
Профессиональный Документы
Культура Документы
Manuscript Draft
Manuscript Number:
Keywords: Hyperspectral; Near Infrared; class-modelling; plastic sorting; hierarchical modelling; flame
retardant
Order of Authors: Jos Manuel Amigo, Ph.D.; Hamid Babamoradi; Saioa Elcoroaristizabal
Aoife Gowen
aoife.gowen@ucd.ie
Expert in Hyperspectral Analysis
Davide Ballabio
davide.ballabio@unimib.it
Expert in class-modelling
Marina Cocchi
marina.cocchi@unimore.it
Expert in class-modelling
Cover Letter(including Novelty Statement)
FACULTY OF SCIENCE
UNIVERSITY OF COPENHAGEN
Dear Editor,
Please find enclosed a copy of the following manuscript to be considered for publication in
Analytica Chimica Acta:
The article is original, unpublished and not being considered for publication elsewhere. We believe
that the manuscript meets the requirements for its publication in Analytica Chimica Acta.
Important to note is that this is a tutorial paper that Prof. Manuel Mir Llad personally
invited us to write.
Sincerely,
Highlights:
- Comprehensive tutorial of Hyperspectral Image analysis
- Hierarchical discrimination of six classes of plastics containing flame retardant
- Step by step guidelines to perform class-modeling on hyperspectral images
- Fusion of multivariate data analysis and digital image processing methods
- Promising methodology for real-time detection of plastics containing flame retardant
*Manuscript (including figures, tables, text graphics and associated captions)
Click here to view linked References
31
32 PS09 0 0 0 0 0 0 0 0 0 0 0 0 16 0
33
34 PS 0 0 0 0 0 0 0 0 0 0 0 0 0 16
35
36
37
38
39
40
41
42
43 8 Remarks and final comments
44
45
46
47
48 This tutorial gives some ideas of how to deal with several issues concerning the development of
49
50 classification models in hyperspectral imaging in a practical and visual manner. In addition, we
51
52
53
have provided with key references and some hints of very fundamental aspects dealing with
54
55 hyperspectral image analysis. As stated at the beginning, every single case in hyperspectral
56
57
58
59
60
61
62 30
63
64
65
1 modelling needs a tailored solution. Therefore, the methods employed in every step must be taken
2
3
4 as an optimal solution for this particular case.
5
6
7
8 Going through the results obtained, all the pellets were correctly classified between the different
9
10
11 main categories of plastics. Nevertheless, there are still some issues regarding the classification of
12
13 the pellets within the classes with and without flame retardant. This is mainly due to the lack of
14
15
16 difference between the pellets of the same plastic, but with different amount of flame retardant.
17
18 From a more applied point of view, the methodology proposed can be easily implemented in a
19
20
21 company for plastic sorting, since all the methods used herein can be easily automatized. Of course,
22
23 there is always room for improvement. To improve the classification results, the first action would
24
25
26
comprise a better separation of the pellets in a, for instance, a conveyor belt. The second action
27
28 would concern looking for increasing the wavelength range. Nowadays is very common to find NIR
29
30 cameras that work between 1000 and 2700 nm, for instance.
31
32
33
34
35 Finally, but not least important, we would like to remark that everyone should look for the
36
37
38 hyperspectral application and multivariate/digital image techniques that better suit to their purposes.
39
40 There are many commercial software and libraries that can be used, as well as many types of
41
42
43
cameras. Therefore, first, think in the application, aim of the analysis and the scientific problem and
44
45 answer yourself these questions:
46
47 - Do we really need hyperspectral imaging? To our knowledge, there are numerous applications
48
49
50 where the use of hyperspectral imaging is not totally necessary. Of course, the capability of fast
51
52 scanning is very attractive to obtain many spectra of a sample in short time. Nevertheless, this does
53
54
55 not justify the use of hyperspectral. Hyperspectral imaging should be used when the
56
57 characterization of the surface of a sample is required.
58
59
60
61
62 31
63
64
65
1 - Which kind of spectral radiation do I need? Obviously, the NIR radiation is the workhorse in
2
3
4 many applications. Nevertheless, there are many cases in which visible imaging can be more than
5
6 enough.
7
8 - Which spatial resolution do I need? This depends on the type of sample to be measured. It is clear
9
10
11 that to characterize the distribution of an active pharmaceutical ingredient in a tablet, a spatial
12
13 resolution in micrometers is needed. But if the samples are big, we might only need spatial
14
15
16 resolutions of millimeters.
17
18 - Which spectral resolution do I need? There are many cases where the target can be achieved by
19
20
21 using two or three selective wavelengths. In this case, filter-based approaches might be wiser than
22
23 scanning a whole spectrum.
24
25
26
- Which models should I use? This is extremely important to assure the selection of the proper data
27
28 analysis tool. Moreover, in recent publications we have seen a trend of using non-linear methods by
29
30 default (e.g. support vector machines SVM). Nevertheless, as personal opinion, we prefer to keep
31
32
33 the analysis as simpler as possible. That is, if your problem can be solved by linear models, there is
34
35 no need to go for non-linear modelling.
36
37
38 - Which kind of software do I need? As explained before, this strongly depends on our ability with
39
40 handling big data. If punctual analysis wants to be performed, we might think in using commercial
41
42
43
software. Nevertheless, if our aim is to master in hyperspectral analysis we strongly advice of using
44
45 software that allows us to program our own routines.
46
47
48
49
50 9 Acknowledgements
51
52 This manuscript has been possible thanks to the financial support received from INNOSORT
53
54 project. INNOSORT is a consortium that brings together technology suppliers, research and
55
56
57
58
59
60
61
62 32
63
64
65
1 knowledge institutions and recipients of waste separation techniques and separated materials. See
2
3
4 further information about INNOSORT and partners in http://innosort.teknologisk.dk/.
5
6
7
8 10 References
9
10
11
[1] R.A. Schowengerdt, Remote Sensing, Third Edition: Models and Methods for Image
12
13 Processing, Elsevier2006.
14
15 [2] E. Adam, O. Mutanga, D. Rugege, Multispectral and hyperspectral remote sensing for
16
17
18 identification and mapping of wetland vegetation: a review, Wetlands Ecology and Management, 18
19
20 (2010) 281-296.
21
22
23 [3] C.A. Lee, S.D. Gasster, A. Plaza, C.I. Chang, B. Huang, Recent Developments in High
24
25 Performance Computing for Remote Sensing: A Review, Ieee Journal of Selected Topics in Applied
26
27
28 Earth Observations and Remote Sensing, 4 (2011) 508-527.
29
30 [4] F.D. van der Meer, H.M.A. van der Werff, F.J.A. van Ruitenbeek, C.A. Hecker, W.H. Bakker,
31
32
33
M.F. Noomen, M. van der Meijde, E.J.M. Carranza, J.B. de Smeth, T. Woldai, Multi- and
34
35 hyperspectral geologic remote sensing: A review, International Journal of Applied Earth
36
37 Observation and Geoinformation, 14 (2012) 112-128.
38
39
40 [5] J.M. Amigo, J. Cruz, M. Bautista, S. Maspoch, J. Coello, M. Blanco, Study of pharmaceutical
41
42 samples by NIR chemical-image and multivariate analysis, Trac-Trend Anal Chem, 27 (2008) 696-
43
44
45 713.
46
47 [6] C. Gendrin, Y. Roggo, C. Collet, Pharmaceutical applications of vibrational chemical imaging
48
49
50
and chemometrics: a review, J Pharm Biomed Anal, 48 (2008) 533-553.
51
52 [7] A.A. Gowen, J.M. Amigo, Applications of Spectroscopy and Chemical Imaging in
53
54 Pharmaceutics, in: J. Popp, V.V. Tuchin, A. Chiou, S.H. Heinemann (Eds.) Handbook of
55
56
57
58
59
60
61
62 33
63
64
65
1 Biophotonics, Volume 3: Photonics in Pharmaceutics, Bioanalysis and Environmental Research,
2
3
4 Wiley2012.
5
6 [8] P.Y. Sacr, C. De Bleye, P.F. Chavez, L. Netchacovitch, P. Hubert, E. Ziemons, Data
7
8 processing of vibrational chemical imaging for pharmaceutical applications, Journal of
9
10
11 Pharmaceutical and Biomedical Analysis, 101 (2014) 123-140.
12
13 [9] J.M. Amigo, Practical issues of hyperspectral imaging analysis of solid dosage forms, Anal
14
15
16 Bioanal Chem, 398 (2010) 93-109.
17
18 [10] Y. Roggo, A. Edmond, P. Chalus, M. Ulmschneider, Infrared hyperspectral imaging for
19
20
21 qualitative analysis of pharmaceutical solid forms, Anal Chim Acta, 535 (2005) 79-87.
22
23 [11] C. Cairs, J.M. Amigo, R. Watt, J. Coello, S. Maspoch, Implementation of enhanced
24
25
26
correlation maps in near infrared chemical images: Application in pharmaceutical research, Talanta,
27
28 79 (2009) 657-664.
29
30 [12] S. Sasic, D.A. Clark, J.C. Mitchell, M.J. Snowden, Raman line mapping as a fast method for
31
32
33 analyzing pharmaceutical bead formulations, Analyst, 130 (2005) 1530-1536.
34
35 [13] M. Khorasani, J.M. Amigo, J. Sonnergaard, P. Olsen, P. Bertelsen, J. Rantanen, Visualization
36
37
38 and prediction of porosity in roller compacted ribbons with near-infrared chemical imaging (NIR-
39
40 CI), Journal of Pharmaceutical and Biomedical Analysis, 109 (2015) 11-17.
41
42
43
[14] C. Ravn, E. Skibsted, R. Bro, Near-infrared chemical imaging (NIR-CI) on pharmaceutical
44
45 solid dosage forms-comparing common calibration approaches, J Pharm Biomed Anal, 48 (2008)
46
47 554-561.
48
49
50 [15] J. Cruz, M. Bautista, J.M. Amigo, M. Blanco, Nir-chemical imaging study of acetylsalicylic
51
52 acid in commercial tablets, Talanta, 80 (2009) 473-478.
53
54
55
56
57
58
59
60
61
62 34
63
64
65
1 [16] J.M. Amigo, I. Mart, A. Gowen, Hyperspectral Imaging and Chemometrics. A Perfect
2
3
4 Combination for the Analysis of Food Structure, Composition and Quality, Data Handling in
5
6 Science and Technology, 2013, pp. 343-370.
7
8 [17] G. ElMasry, M. Kamruzzaman, D.W. Sun, P. Allen, Principles and Applications of
9
10
11 Hyperspectral Imaging in Quality Evaluation of Agro-Food Products: A Review, Critical Reviews
12
13 in Food Science and Nutrition, 52 (2012) 999-1023.
14
15
16 [18] G. Elmasry, N. Wang, A. ElSayed, M. Ngadi, Hyperspectral imaging for nondestructive
17
18 determination of some quality attributes for strawberry, J Food Eng, 81 (2007) 98-107.
19
20
21 [19] P.M. Mehl, Y.-R. Chen, M.S. Kim, D.E. Chan, Development of hyperspectral imaging
22
23 technique for the detection of apple surface defects and contaminations, J Food Eng, 61 (2004) 67-
24
25
26
81.
27
28 [20] D. Lorente, N. Aleixos, J. Gomez-Sanchis, S. Cubero, O.L. Garcia-Navarrete, J. Blasco,
29
30 Recent Advances and Applications of Hyperspectral Imaging for Fruit and Vegetable Quality
31
32
33 Assessment, Food and Bioprocess Technology, 5 (2012) 1121-1142.
34
35 [21] D. Sun, Hyperspectral Imaging for Food Quality Analysis and Control, Elsevier Inc.2010.
36
37
38 [22] A.A. Gowen, C.P. O'Donnell, P.J. Cullen, G. Downey, J.M. Frias, Hyperspectral imaging - an
39
40 emerging process analytical tool for food quality and safety control, Trends in Food Science &
41
42
43
Technology, 18 (2007) 590-598.
44
45 [23] J.H. Cheng, D.W. Sun, Hyperspectral imaging as an effective tool for quality analysis and
46
47 control of fish and other seafoods: Current research and potential applications, Trends in Food
48
49
50 Science and Technology, 37 (2014) 78-91.
51
52 [24] S. Cubero, N. Aleixos, E. Molt, J. Gmez-Sanchis, J. Blasco, Advances in Machine Vision
53
54
55 Applications for Automatic Inspection and Quality Evaluation of Fruits and Vegetables, Food and
56
57 Bioprocess Technology, 4 (2011) 487-504.
58
59
60
61
62 35
63
64
65
1 [25] I. Kim, M.S. Kim, Y.R. Chen, S.G. Kong, Detection of skin tumors on chicken carcasses using
2
3
4 hyperspectral fluorescence imaging, Transactions of the Asae, 47 (2004) 1785-1792.
5
6 [26] Y.Z. Feng, D.W. Sun, Application of Hyperspectral Imaging in Food Safety Inspection and
7
8 Control: A Review, Critical Reviews in Food Science and Nutrition, 52 (2012) 1039-1058.
9
10
11 [27] K. Flynn, R. O'Leary, C. Roux, B.J. Reedy, Forensic analysis of bicomponent fibers using
12
13 infrared chemical imaging, Journal of forensic sciences, 51 (2006) 586-596.
14
15
16 [28] G.J. Edelman, E. Gaston, T.G. van Leeuwen, P.J. Cullen, M.C.G. Aalders, Hyperspectral
17
18 imaging for non-contact analysis of forensic traces, Forensic Sci Int, 223 (2012) 28-39.
19
20
21 [29] M.A.F. De La Ossa, C. Garca-Ruiz, J.M. Amigo, Near infrared spectral imaging for the
22
23 analysis of dynamite residues on human handprints, Talanta, 130 (2014) 315-321.
24
25
26
[30] M.T. Fernndez de la Ossa, J.M. Amigo, C. Garca-Ruiz, Detection of residues from explosive
27
28 manipulation by near infrared hyperspectral imaging: A promising forensic tool, Forensic Sci Int,
29
30 242 (2014) 228-235.
31
32
33 [31] C.S. Silva, M.F. Pimentel, R.S. Honorato, C. Pasquini, J.M. Prats-Montalban, A. Ferrer, Near
34
35 infrared hyperspectral imaging for forensic analysis of document forgery, Analyst, 139 (2014)
36
37
38 5176-5184.
39
40 [32] K. Flynn, R. O'Leary, C. Lennard, C. Roux, B.J. Reedy, Forensic applications of infrared
41
42
43
chemical imaging: Multi-layered paint chips, Journal of forensic sciences, 50 (2005) 832-841.
44
45 [33] G.L. Lu, B.W. Fei, Medical hyperspectral imaging: a review, J Biomed Opt, 19 (2014).
46
47 [34] R.F. Kokaly, G.P. Asner, S.V. Ollinger, M.E. Martin, C.A. Wessman, Characterizing canopy
48
49
50 biochemistry from imaging spectroscopy and its application to ecosystem studies, Remote Sens
51
52 Environ, 113 (2009) S78-S91.
53
54
55
56
57
58
59
60
61
62 36
63
64
65
1 [35] C.G. Zhang, H. Wang, J.F. Huang, Q. Gao, The visible to the near infrared narrow band
2
3
4 acousto-optic tunable filter and the hyperspectral microscopic imaging on biomedicine study, J
5
6 Optics-Uk, 16 (2014).
7
8 [36] M.E. Klein, B.J. Aalderink, R. Padoan, G. de Bruin, T.A.G. Steemers, Quantitative
9
10
11 hyperspectral reflectance imaging, Sensors, 8 (2008) 5576-5618.
12
13 [37] F. Rosi, C. Miliani, R. Braun, R. Harig, D. Sali, B.G. Brunetti, A. Sgamellotti, Noninvasive
14
15
16 Analysis of Paintings by Mid-infrared Hyperspectral Imaging, Angew Chem Int Edit, 52 (2013)
17
18 5258-5261.
19
20
21 [38] Z.M. Xu, E.Y. Lam, Image reconstruction using spectroscopic and hyperspectral information
22
23 for compressive terahertz imaging, J Opt Soc Am A, 27 (2010) 1638-1646.
24
25
26
[39] C.M. Wang, S.C. Yang, P.C. Chung, C.I. Chang, C.S. Lo, C.C. Chen, C.W. Yang, C.H. Wen,
27
28 Orthogonal subspace projection-based approaches to classification of MR image sequences,
29
30 Comput Med Imag Grap, 25 (2001) 465-476.
31
32
33 [40] S. Sasic, Y. Ozaki, Raman, Infrared, and Near-Infrared Chemical Imaging, Wiley2010.
34
35 [41] J.A.F. Pierna, P. Vermeulen, O. Amand, A. Tossens, P. Dardenne, V. Baeten, NIR
36
37
38 hyperspectral imaging spectroscopy and chemometrics for the detection of undesirable substances
39
40 in food and feed, Chemometr Intell Lab, 117 (2012) 233-239.
41
42
43
[42] P. Geladi, H. Grahn, M. Manley, Data Analysis and Chemometrics for Hyperspectral Imaging,
44
45 in: S. Sasic, Y. Ozaki (Eds.) Raman, Infrared, and Near-Infrared Chemical Imaging, Wiley2010.
46
47 [43] C.I. Chang, X. Jiao, C.C. Wu, Y. Du, M.L. Chang, A review of unsupervised spectral target
48
49
50 analysis for hyperspectral imagery, Eurasip Journal on Advances in Signal Processing, 2010 (2010).
51
52 [44] D.M. Haaland, H.D.T. Jones, M.H. Van Benthem, M.B. Sinclair, D.K. Melgaard, C.L. Stork,
53
54
55 M.C. Pedroso, P. Liu, A.R. Brasier, N.L. Andrews, D.S. Lidke, Hyperspectral Confocal
56
57
58
59
60
61
62 37
63
64
65
1 Fluorescence Imaging: Exploring Alternative Multivariate Curve Resolution Approaches, Applied
2
3
4 Spectroscopy, 63 (2009) 271-279.
5
6 [45] J.M. Prats-Montalbn, A. de Juan, A. Ferrer, Multivariate image analysis: A review with
7
8 applications, Chemometr Intell Lab, 107 (2011) 1-23.
9
10
11 [46] B. Boldrini, W. Kessler, K. Rebnera, R.W. Kessler, Hyperspectral imaging: A review of best
12
13 practice, performance and pitfalls for in-line and on-line applications, Journal of Near Infrared
14
15
16 Spectroscopy, 20 (2012) 483-508.
17
18 [47] M. Vidal, J.M. Amigo, Pre-processing of hyperspectral images. Essential steps before image
19
20
21 analysis, Chemometr Intell Lab, 117 (2012) 138-148.
22
23 [48] G. Bonifazi, L. Damiani, S. Serranti, E.J. Bakker, P.C. Rem, Innovative Sensing Technologies
24
25
26
Applied to Post-Consumer Polyolefins Recovery, Metal Int, 14 (2009) 5-10.
27
28 [49] S. Serranti, A. Gargiulo, G. Bonifazi, Characterization of post-consumer polyolefin wastes by
29
30 hyperspectral imaging for quality control in recycling processes, Waste Manage, 31 (2011) 2217-
31
32
33 2227.
34
35 [50] S. Serranti, A. Gargiulo, G. Bonifazi, Classification of polyolefins from building and
36
37
38 construction waste using NIR hyperspectral imaging system, Resour Conserv Recy, 61 (2012) 52-
39
40 58.
41
42
43
[51] A. Ulrici, S. Serranti, C. Ferrari, D. Cesare, G. Foca, G. Bonifazi, Efficient chemometric
44
45 strategies for PET-PLA discrimination in recycling plants using hyperspectral imaging, Chemometr
46
47 Intell Lab, 122 (2013) 31-39.
48
49
50 [52] C.A. de Wit, An overview of brominated flame retardants in the environment, Chemosphere,
51
52 46 (2002) 583-624.
53
54
55 [53] Shimadtzu, Identification of brominated flame retardants in polymers, Shimadtzu News,
56
57 Shimatzu, 2005, pp. 10-11.
58
59
60
61
62 38
63
64
65
1 [54] S. Kikuchi, K. Kawauchi, S. Ooki, M. Kurosawa, H. Honjho, T. Yagishita, Non-destructive
2
3
4 rapid analysis of brominated flame retardants in electrical and electronic equipment using Raman
5
6 spectroscopy, Anal Sci, 20 (2004) 1111-1112.
7
8 [55] M. Stepputat, R. Noll, On-line detection of heavy metals and brominated flame retardants in
9
10
11 technical polymers with laser-induced breakdown spectrometry, Applied Optics, 42 (2003) 6210-
12
13 6220.
14
15
16 [56] A. Golloch, D. Siegmund, Sliding spark spectroscopy - rapid survey analysis of flame
17
18 retardants and other additives in polymers, Fresen J Anal Chem, 358 (1997) 804-811.
19
20
21 [57] C. Gallen, A. Banks, S. Brandsma, C. Baduel, P. Thai, G. Eaglesham, A. Heffernan, P.
22
23 Leonards, P. Bainton, J.F. Mueller, Towards development of a rapid and effective non-destructive
24
25
26
testing strategy to identify brominated flame retardants in the plastics of consumer products,
27
28 Science of the Total Environment, 491 (2014) 255-265.
29
30 [58] M. Schlummer, F. Brandl, A. Maurer, R. van Eldik, Analysis of flame retardant additives in
31
32
33 polymer fractions of waste of electric and electronic equipment (WEEE) by means of HPLC-
34
35 UV/MS and GPC-HPLC-UV, J Chromatogr A, 1064 (2005) 39-51.
36
37
38 [59] J. Burger, P. Geladi, Hyperspectral NIR image regression part II: Dataset preprocessing
39
40 diagnostics, J Chemometr, 20 (2006) 106-119.
41
42
43
[60] J. Burger, P. Geladi, Hyperspectral NIR image regression part I: calibration and correction, J
44
45 Chemometr, 19 (2005) 355-363.
46
47 [61] P. Geladi, J. Burger, T. Lestander, Hyperspectral imaging: calibration problems and solutions,
48
49
50 Chemometr Intell Lab, 72 (2004) 209-217.
51
52 [62] Phyton, https://www.python.org/.
53
54
55 [63] RStudio, http://www.rstudio.com/.
56
57 [64] I. The MathWorks, http://se.mathworks.com/.
58
59
60
61
62 39
63
64
65
1 [65] M.T.b.E.R. Inc., http://www.eigenvector.com/software/mia_toolbox.htm.
2
3
4 [66] ENVI, http://www.exelisvis.com/ProductsServices/ENVIProducts/ENVI.aspx.
5
6 [67] J.M. Amigo, C. Ravn, N.B. Gallagher, R. Bro, A comparison of a common approach to partial
7
8 least squares-discriminant analysis and classical least squares in hyperspectral imaging, Int J
9
10
11 Pharmaceut, 373 (2009) 179-182.
12
13 [68] E.R. Inc., http://eigenvector.com/.
14
15
16 [69] I. The MathWorks, Image Processing Toolbox, 2015.
17
18 [70] F.M. Mirabella, Internal-Reflection Spectroscopy, Appl Spectrosc Rev, 21 (1985) 45-178.
19
20
21 [71] A. Rinnan, F. van den Berg, S.B. Engelsen, Review of the most common pre-processing
22
23 techniques for near-infrared spectra, Trac-Trend Anal Chem, 28 (2009) 1201-1222.
24
25
26
[72] M. Barker, W. Rayens, Partial least squares for discrimination, J Chemometr, 17 (2003) 166-
27
28 173.
29
30 [73] I.G. Chong, C.H. Jun, Performance of some variable selection methods when multicollinearity
31
32
33 is present, Chemometr Intell Lab, 78 (2005) 103-112.
34
35 [74] D. Lorente, J. Blasco, A.J. Serrano, E. Soria-Olivas, N. Aleixos, J. Gomez-Sanchis,
36
37
38 Comparison of ROC Feature Selection Method for the Detection of Decay in Citrus Fruit Using
39
40 Hyperspectral Images, Food and Bioprocess Technology, 6 (2013) 3613-3619.
41
42
43
[75] J. Demsar, Statistical comparisons of classifiers over multiple data sets, J Mach Learn Res, 7
44
45 (2006) 1-30.
46
47 [76] D.J. Hand, Assessing the Performance of Classification Methods, Int Stat Rev, 80 (2012) 400-
48
49
50 414.
51
52 [77] R.C. Gonzalez, R.E. Woods, E.L. Eddins, Digital Image Processing Using MATLAB,
53
54
55 Gatesmark Publishing2009.
56
57
58
59
60
61
62 40
63
64
65
1 [78] J. Canny, A Computational Approach to Edge-Detection, Ieee T Pattern Anal, 8 (1986) 679-
2
3
4 698.
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62 41
63
64
65
Tables
Table 1. Sensitivity, specificity ad classification error for the five main categories of plastics
GLOBAL MODEL
PA6 PBT PP PS ABS
Sensitivity (CAL) 0,945 0,993 0,964 0,991 0,985
Sensitivity (CV) 0,946 0,993 0,963 0,991 0,983
Sensitivity (PRED) 0,864 0,996 0,967 0,983 0,972
ABS PS
ABS10 ABS15 ABS02 ABS PS00 PS07 PS09 PS
Sensitivity (CAL) 0,892 0,87 0,831 0,755 0,931 0,878 0,698 0,81
Sensitivity (CV) 0,891 0,87 0,826 0,752 0,924 0,873 0,693 0,812
Specificity (CAL) 0,57 0,837 0,737 0,635 0,83 0,763 0,65 0,842
Specificity (CV) 0,569 0,837 0,737 0,631 0,83 0,762 0,648 0,841
Class error (CAL) 0,269 0,146 0,215 0,304 0,119 0,179 0,325 0,174
Class error (CV) 0,27 0,146 0,218 0,308 0,122 0,182 0,329 0,173
PBT PA6 PP
PBT010 PBT PA6010 PA6 PP202 PP
Sensitivity (CAL) 0,807 0,599 0,856 0,893 0,945 0,944
Sensitivity (CV) 0,798 0,603 0,852 0,895 0,944 0,945
Predicted class
ABS010 ABS015 ABS020 ABS PA6010 PA6 PBT010 PBT PP202 PP PS00 PS07 PS09 PS
ABS010 2 0 7 1 0 0 0 0 0 0 0 0 0 0
ABS015 0 11 0 9 0 0 0 0 0 0 0 0 0 0
ABS020 0 0 12 0 0 0 0 0 0 0 0 0 0 0
ABS 0 0 0 21 0 0 0 0 0 0 0 0 0 0
PA6010 0 0 0 0 8 0 0 0 0 0 0 0 0 0
PA6 0 0 0 0 0 22 0 0 0 0 0 0 0 0
Real class
PBT010 0 0 0 0 0 0 12 5 0 0 0 0 0 0
PBT 0 0 0 0 0 0 0 13 0 0 0 0 0 0
PP202 0 0 0 0 0 0 0 0 11 0 0 0 0 0
PP 0 0 0 0 0 0 0 0 0 16 0 0 0 0
PS00 0 0 0 0 0 0 0 0 0 0 8 0 0 4
PS07 0 0 0 0 0 0 0 0 0 0 0 8 0 4
PS09 0 0 0 0 0 0 0 0 0 0 0 0 16 0
PS 0 0 0 0 0 0 0 0 0 0 0 0 0 16
Figure 01
Click here to download high resolution image
Figure 02
Click here to download high resolution image
Figure 03
Click here to download high resolution image
Figure 04
Click here to download high resolution image
Figure 05
Click here to download high resolution image
Figure 06
Click here to download high resolution image
Figure 07
Click here to download high resolution image
Figure 08
Click here to download high resolution image
Figure 09
Click here to download high resolution image
Figure 10
Click here to download high resolution image
Figure 11
Click here to download high resolution image
Figure 12
Click here to download high resolution image
JMAmigo