Documente Academic
Documente Profesional
Documente Cultură
Original papers
a r t i c l e i n f o a b s t r a c t
Article history: This paper evaluates the feasibility of laser-induced backscattering imaging for the classification of
Received 27 October 2016 seeded and seedless watermelons during storage. Backscattering images were obtained from seeded
Received in revised form 20 April 2017 and seedless watermelon samples through a laser diode emitting at 658 nm using a backscattering imag-
Accepted 10 June 2017
ing system developed for the purpose. The pre-processed datasets extracted from the backscattering
Available online 20 June 2017
images were analysed using principal component analysis (PCA). The datasets were separated into train-
ing (75%) and testing (25%) datasets as the inputs in the classification algorithms. Three multivariate pat-
Keywords:
tern recognition algorithms were used including linear discriminant analysis (LDA), quadratic
Backscattering imaging
Laser light
discriminant analysis (QDA), and k-nearest neighbour (kNN). The QDA-based algorithms obtained the
Watermelon highest overall average classification accuracies (100%) for both the seeded and seedless watermelons.
Pattern recognition algorithm The LDA and kNN-based algorithms also obtained quite high classification accuracies with all the accu-
Storage racies above 90%. The laser-induced backscattering imaging technique is potentially useful for classifica-
tion of seeded and seedless watermelons.
Ó 2017 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.compag.2017.06.010
0168-1699/Ó 2017 Elsevier B.V. All rights reserved.
312 M. Mohd Ali et al. / Computers and Electronics in Agriculture 140 (2017) 311–316
also one of the criteria as the shape and size vary between both and 80–90% RH). The experiments were carried out in two batches
varieties (Zheng et al., 2006). Conversely, seedless watermelon is for three weeks in order to identify the similarity and differentia-
grown to have immature seed which is developed from the triploid tion throughout the storage period. For each cultivar, 70 Black
trait and comprises of numerous large seeds throughout the flesh Beauty and Red Seedless watermelons were measured with a total
of the hybrid watermelon. For this reason, the seedless variety is of 140 watermelons for each batch, respectively. Seven interval
more valuable in commercial markets due to the preference and storage days were used to analyse the physicochemical changes
convenience requirements of the customers. of both varieties (day 0, day 4, day 8, day 12, day 15, day 18, and
Recently, laser-induced backscattering imaging has been day 21). Ten watermelons were randomly sampled from both the
applied for non-destructively evaluating various agricultural pro- seeded and seedless watermelons for each storage day under the
duce in terms of food quality and control as an alternative to the same storage condition. The same fruits were also used for both
machine vision method. Previous works have discussed the feasi- backscattering acquisition and destructive measurement.
bility of laser-induced backscattering imaging in evaluating the
food quality of fruits and vegetables such as apples (Qing et al., 2.2. Imaging system
2007; Romano et al., 2011), bell peppers (Romano et al., 2012),
tomatoes (Mollazade et al., 2013; Tu et al., 2000), bananas In this study, a laser-induced backscattering imaging system
(Adebayo et al., 2016b; Hashim et al., 2014, 2013, 2012), kiwifruits which consists of a charge-coupled device (CCD) camera (QICAM
(Baranyai and Zude, 2009), citrus (Lorente et al., 2015; Lorente Colour Fast 1394, QImaging, Surrey, BC, Canada) with a zoom lens
et al., 2013), and papayas (Udomkun et al., 2014). Romano et al. (F5.6 and focal length of 18 mm), a laser diode emitting at 658 nm
(2011) applied laser light backscattering imaging emitted in the with 30 mW maximal power, and a computer equipped with the
visible wavelength at 635 nm to determine moisture content, sol- Image-Pro Insight 9 software (Media Cybernetics, Inc., USA) was
uble solids content, and hardness of apples during drying. developed for acquiring the backscattering images. The CCD cam-
Lorente et al. (2013) used a laser diode emitting at five different era captured backscattering images of 1392 1040 pixels with a
wavelengths (532, 660, 785, 830, and 1060 nm) to detect decay resolution of 0.073 mm/pixel and 12-bit gray colour depth. The
in citrus fruits in comparison with respective laser wavelengths. portion of backscattering light upon penetrating into the fruit tis-
In previous research, Hashim et al. (2013) investigated the effect sue was recorded by the CCD camera and stored on the computer.
of water content, fruit pigment, and chilling injury in bananas with A frame (930 mm 700 mm 700 mm) with a lightproof medium
respect to storage time and temperature using laser-induced i.e. black cloth was used to avoid interruption from ambient light.
backscattering imaging. Dénes et al. (2013) reported the applica- A laser diode at 658 nm is used as the light source which can dis-
tion of a backscattering imaging system on the optical properties tribute light for a selected wavelength. Since the processing of the
of bananas based on the influence of different drying times, tem- whole laser-induced backscattering imaging system is quite fast,
peratures, and pre-treatments emitting at seven wavelengths the image acquisition was set at a short exposure time (10 ms).
(532, 635, 650, 780, 808, 850, and 1064 nm). Qing et al. (2008) The distance from the camera lens to the samples was set at
analysed backscattering images of apples to determine the firm- 55 cm in a fixed position at the top of the fruit. The incident angle
ness and soluble solids content in different growing locations. of the light beam was placed at 22° with respect to the vertical
Basically, laser induced backscattering imaging uses the theory axis. The backscattering images were obtained by placing the sam-
of light propagation when the light is reflected and scattered back ple manually on the sample holder facing the CCD camera.
towards the external surface of the sample upon entering the fruit Backscattering images obtained at 658 nm from seeded and seed-
tissue. In this case, the backscattering photons obtained through less watermelon samples during storage are shown in Fig. 1. Over-
light propagation carry useful information associated with the all, a total of six images were acquired for each of the 140 samples
morphology properties of the flesh when interacting with the qual- of watermelon skin, thereby resulting in 840 backscattering
ity attributes of the fruit sample (Adebayo et al., 2016a; Mollazade images.
et al., 2012). Moreover, the methods developed for processing the
backscattering images are computationally inexpensive. The low 2.3. Backscattering image analysis
processing cost suggests that a laser-based system could possibly
be developed and applied, especially in real-time sorting/grading The backscattering images were obtained through an image
machines. However, as far as is known, no study has reported a segmentation process using Matlab software (Version R2013a,
technique able to classify seeded and seedless watermelons cou- The Mathworks Inc., Natick, MA, USA). The region of interest in
pled with classifier models using a laser-induced backscattering the image was selected by choosing the threshold value from a his-
imaging system. Thus, the ultimate aim of this study is to evaluate togram profile based on the backscattering image (Fig. 2). From the
the application of laser-induced backscattering imaging for the feature extraction of the backscattering image, the pixel value
classification of seeded and seedless watermelons during storage. (maximum intensity, minimum intensity, and mean intensity)
This study also investigates different pattern recognition algorithm and shape (major axis length, minor axis length, and perimeter)
models for the classification of seeded and seedless watermelons. measurements were determined. The backscattering parameter
measurements were then exported to Unscrambler software (Ver-
sion 10.3, CAMO AS, Oslo, Norway) for further statistical analysis.
2. Materials and methods
2.4. Multivariate pattern recognition algorithms
2.1. Sample preparation
Linear discriminant analysis (LDA), quadratic discriminant anal-
Seeded watermelons from the Black Beauty variety and seedless ysis (QDA), and k-nearest neighbour (kNN) algorithms are different
watermelons from the Red Seedless variety were procured from a multivariate pattern recognition algorithms that can be used to
local farm in Manjung, Perak, a middle state in Malaysia. The classify the watermelon cultivars and storage days based on the
watermelons were harvested 65 days after planting and delivered backscattering parameters. These algorithms have been reported
to a commercial retailer (Melon Master Sdn. Bhd.). The fruits were in the literature for different classification purposes, especially to
transported immediately to the Biomaterials Processing Labora- classify spectral data sets (Liaghat et al., 2014; Sankaran et al.,
tory, Universiti Putra Malaysia and stored in cold storage (12 °C 2011). Before determining the overall classification accuracies of
M. Mohd Ali et al. / Computers and Electronics in Agriculture 140 (2017) 311–316 313
Fig. 1. Backscattering images of the fruit samples using laser diode with wavelength of 658 nm for (a) seeded and (b) seedless watermelons.
Fig. 2. Backscattering image analysis: (a) raw backscattering image and (b) histogram profile of pixels of image.
the different algorithms, the ratio of training to testing dataset along with k-value yielded the highest overall and individual class
(70:30) was calculated using the combined datasets. The PC scores for the average classification accuracies and was regarded as the
were randomised and divided into training datasets for developing best model for the kNN algorithm.
the classification model whereas the testing datasets were used to
test the model. In this study, LDA and QDA used two PC scores
obtained from the combined datasets. The kNN model consisted 3. Results and discussion
of two classes which were derived from the training datasets and
predicted by determining the distance between the unknown sam- 3.1. Potential of laser-induced backscattering imaging to classify
ple and k nearest neighbours. All multivariate algorithms were watermelons
developed in the SAS software (Version 9.4, SAS Institute, Cary,
NC, USA). The average classification accuracies were calculated Principal component analysis (PCA) was performed to further
by taking the average of five readings to randomise the data sets. minimise the backscattering parameters into uncorrelated vari-
The overall and individual classes (seeded and seedless) classifica- ables, also known as PC. A clear discrimination between the seeded
tion accuracies were obtained from a confusion matrix. and seedless watermelons by PC1 and PC2 based on the PCA score
Discriminant functions were obtained from the input indepen- plot was observed (Fig. 3). Two PCs including PC1 captured 61% of
dent variables via the training data sets, which were used to deter- the variance and PC2 captured 27% of the variance thus showing a
mine the classification based on the principal component (PC) clear separation between the seeded and seedless watermelons.
score values between the classes. In discriminant analysis, LDA is Based on the results, laser-induced backscattering imaging was
a linear model whereas QDA is known as a quadratic model, which found to be able to provide the necessary classification of seeded
generates the classification of the testing datasets. Apart from the and seedless watermelons. The PC score was randomised and plot-
discriminant analysis algorithms, a kNN algorithm was also used to ted in an effort to classify the classes of correlated variables of the
classify between the classes from the training datasets. The two seeded and seedless watermelons. A preliminary set of analyses
classes of datasets (seeded and seedless) were predicted by calcu- was performed on the samples where 75% of the datasets were
lating the distance between the k nearest neighbours and the used for training the classification algorithm, and 25% of the data-
classes of samples. The optimum k-value (constant) in the algo- sets were used as the testing dataset.
rithm was chosen from the highest classification accuracies during By observing the PCA score plot, it was possible to classify the
the kNN algorithm development. In this case, the kNN algorithm relationship between the backscattering parameters as well as
314 M. Mohd Ali et al. / Computers and Electronics in Agriculture 140 (2017) 311–316
Table 1
The average classification accuracies of seeded and seedless watermelons at different storage days using multivariate pattern recognition algorithms.
Table 2
The classification accuracies of seeded and seedless watermelons at different storage days based on the backscattering parameters.
Storage Cultivar Major axis length Minor axis length Mean intensity Maximum intensity Minimum intensity Perimeter
day
Correct Missed Correct Missed Correct Missed Correct Missed Correct Missed Correct Missed
(%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%) (%)
Day 0 Seeded 70 30 50 50 50 50 80 20 50 50 50 50
Seedless 80 20 80 20 70 30 100 0 70 30 90 10
Day 4 Seeded 70 30 60 40 100 0 100 0 80 20 50 50
Seedless 100 0 40 60 100 0 100 0 80 20 80 20
Day 8 Seeded 50 50 50 50 100 0 90 10 60 40 50 50
Seedless 100 0 100 0 60 40 70 30 100 0 90 10
Day 12 Seeded 60 40 50 50 100 0 100 0 60 40 50 50
Seedless 90 10 80 20 90 10 100 0 80 20 100 0
Day 15 Seeded 80 20 80 20 90 10 80 20 70 30 70 30
Seedless 90 10 90 10 70 30 70 30 100 0 90 10
Day 18 Seeded 80 20 40 60 100 0 100 0 100 0 60 40
Seedless 80 20 70 30 100 0 100 0 50 50 70 30
Day 21 Seeded 100 0 100 0 80 20 40 60 100 0 100 0
Seedless 100 0 100 0 70 30 90 10 100 0 100 0
4. Conclusions
the development of suitable imaging techniques offers new and Lorente, D., Zude, M., Regen, C., Palou, L., Gómez-Sanchis, J., Blasco, J., 2013. Early
decay detection in citrus fruit using laser-light backscattering imaging.
robust devices by means of an automatic inspection system.
Postharvest Biol. Technol. 86, 424–430.
Mohd Ali, M., Hashim, N., Bejo, S.K., Shamsudin, R., 2017. Quality evaluation of
Acknowledgments watermelon using laser-induced backscattering imaging during storage.
Postharvest Biol. Technol. 123, 51–59.
Mollazade, K., Omid, M., Akhlaghian Tab, F., Kalaj, Y.R., Mohtasebi, S.S., Zude, M.,
The authors are thankful for the support received from the 2013. Analysis of texture-based features for predicting mechanical properties of
Department of Biological and Agricultural Engineering, Faculty of horticultural products by laser light backscattering imaging. Comput. Electron.
Engineering, Universiti Putra Malaysia for supplying the facilities Agric. 98, 34–45.
Mollazade, K., Omid, M., Tab, F.A., Mohtasebi, S.S., 2012. Principles and applications
for this research under the Fundamental Research Grant Scheme of light backscattering imaging in quality evaluation of agro-food products: a
(FRGS) (Vot. No.: 5524604). review. Food Bioprocess Technol. 5, 1465–1485.
Perkins-Veazie, P., Collins, J.K., 2004. Flesh quality and lycopene stability of fresh-
cut watermelon. Postharvest Biol. Technol. 31, 159–166.
References Perkins-Veazie, P., Collins, J.K., Davis, A.R., Roberts, W., 2006. Carotenoid content of
50 watermelon cultivars. J. Agric. Food Chem. 54, 2593–2597.
Adebayo, S.E., Hashim, N., Abdan, K., Hanafi, M., 2016a. Application and potential of Qing, Z., Ji, B., Zude, M., 2008. Non-destructive analyses of apple quality parameters
backscattering imaging techniques in agricultural and food processing – a by means of laser-induced light backscattering imaging. Postharvest Biol.
review. J. Food Eng. 169, 155–164. Technol. 48, 215–222.
Adebayo, S.E., Hashim, N., Abdan, K., Hanafi, M., Mollazade, K., 2016b. Prediction of Qing, Z., Ji, B., Zude, M., 2007. Predicting soluble solid content and firmness in apple
quality attributes and ripeness classification of bananas using optical fruit by means of laser light backscattering image analysis. J. Food Eng. 82, 58–
properties. Sci. Hortic. (Amsterdam) 212, 171–182. 67.
Ajuru, M.G., Okoli, B.E., 2013. The morphological characterization of the melon Rimando, A.M., Perkins-Veazie, P.M., 2005. Determination of citrulline in
species in the family Cucurbitaceae Juss., and their utilization in Nigeria. Int. J. watermelon rind. J. Chromatogr. A 1078, 196–200.
Mod. Bot. 3, 15–19. Romano, G., Argyropoulos, D., Nagle, M., Khan, M.T., Müller, J., 2012. Combination of
Baranyai, L., Zude, M., 2009. Analysis of laser light propagation in kiwifruit using digital images and laser light to predict moisture content and color of bell
backscattering imaging and Monte Carlo simulation. Comput. Electron. Agric. pepper simultaneously during drying. J. Food Eng. 109, 438–448.
69, 33–39. Romano, G., Nagle, M., Argyropoulos, D., Müller, J., 2011. Laser light backscattering
Dénes, D.L., Parrag, V., Felföldi, J., Baranyai, L., 2013. Influence of parameters of to monitor moisture content, soluble solid content and hardness of apple tissue
drying on laser induced diffuse reflectance of banana discs. J. Food Phys. 26, 11– during drying. J. Food Eng. 104, 657–662.
16. Sankaran, S., Mishra, A., Maja, J.M., Ehsani, R., 2011. Visible-near infrared
FAO, 2015. Global production of vegetables in 2013, by type (in million metric tons). spectroscopy for detection of Huanglongbing in citrus orchards. Comput.
Retrieved from <http://www.statista.com/statistics/264065/global-production- Electron. Agric. 77, 127–134.
of-vegetables-by-type/> (accessed 24.04.16). Tlili, I., Hdider, C., Lenucci, M.S., Riadh, I., Jebari, H., Dalessandro, G., 2011. Bioactive
Griffiths, A.J., Gelbart, W.M., Miller, J.H., Lewontin, R.C., 1999. Changes in compounds and antioxidant activities of different watermelon (Citrullus lanatus
Chromosome Number. W. H. Freeman, New York. (Thunb.) Mansfeld) cultivars as affected by fruit sampling area. J. Food Compos.
Hashim, N., Janius, R.B., Abdul, R., Osman, A., Shitan, M., Zude, M., 2014. Changes of Anal. 24, 307–314.
backscattering parameters during chilling injury in bananas. J. Eng. Sci. Technol. Tu, K., Jancsok, P., Nicolai, B., Baerdemaeker, J..De., 2000. Use of laser-scattering
9, 314–325. imaging to study tomato-fruit quality in relation to acoustic and compression
Hashim, N., Janius, R.B., Baranyai, L., Rahman, R.A., Osman, A., Zude, M., 2012. measurements. Int. J. Food Sci. Technol. 35, 503.
Kinetic model for colour changes in bananas during the appearance of chilling Udomkun, P., Nagle, M., Mahayothee, B., Müller, J., 2014. Laser-based imaging
injury symptoms. Food Bioprocess Technol. 5, 2952–2963. system for non-invasive monitoring of quality changes of papaya during drying.
Hashim, N., Pflanz, M., Regen, C., Janius, R.B., Abdul Rahman, R., Osman, A., Shitan, Food Control 42, 225–233.
M., Zude, M., 2013. An approach for monitoring the chilling injury appearance Wechter, W.P., Levi, A., Harris, K.R., Davis, A.R., Fei, Z., Katzir, N., Giovannoni, J.J.,
in bananas by means of backscattering imaging. J. Food Eng. 116, 28–36. Salman-Minkov, A., Hernandez, A., Thimmapuram, J., Tadmor, Y., Portnoy, V.,
Liaghat, S., Mansor, S., Ehsani, R., Shafri, H.Z.M., Meon, S., Sankaran, S., 2014. Mid- Trebitsh, T., 2008. Gene expression in developing watermelon fruit. BMC
infrared spectroscopy for early detection of basal stem rot disease in oil palm. Genomics 9, 275–287.
Comput. Electron. Agric. 101, 48–54. Zhang, B., Huang, W., Li, J., Zhao, C., Fan, S., Wu, J., Liu, C., 2014. Principles,
Liu, W., Zhao, S., Cheng, Z., Wan, X., Yan, Z., King, S.R., 2010. Lycopene and citrulline developments and applications of computer vision for external quality
contents in watermelon (Citrullus lanatus) fruit with different ploidy and inspection of fruits and vegetables: a review. Food Res. Int. 62, 326–343.
changes during fruit development. Acta Hortic. 871, 543–550. Zheng, C., Sun, D.-W., Zheng, L., 2006. Recent developments and applications of
Lorente, D., Zude, M., Idler, C., Gómez-Sanchis, J., Blasco, J., 2015. Laser-light image features for food quality evaluation and inspection – a review. Trends
backscattering imaging for early decay detection in citrus fruit using both a Food Sci. Technol. 17, 642–655.
statistical and a physical model. J. Food Eng. 154, 76–85.