Ukrainian Journal of Physical Optics


2022 Volume 23, Issue 4


ISSN 1816-2002 (Online), ISSN 1609-1833 (Print)

Hazelnut quality detection based on deep learning and near-infrared spectroscopy

Dandan Li, Dongyan Zhang, Dapeng Jiang, Jun Cao and Jiuqing Liu

College of Mechanical and Electrical Engineering, Northeast Forestry University, Harbin 150040, Heilongjiang, China

ABSTRACT

Hazelnut kernels can often be incomplete or malformed and have some other defects. Manual methods for their classification and recognition or even old-fashioned machine-learning approaches are prone to the problems of low recognition efficiency and high misjudgement rates. In this study we apply deep learning to the problem of recognition of hazelnut-kernel defects. Two convolution neural-network systems, MobileNetV2 and Resnet-50, are used for training and recognition. It is found that the prediction accuracy of the ResNet-50 training set is improved by 10.3% and the training-loss rate reduced by 0.041, if compared with MobileNetV2. Moreover, the validation accuracy achieved with ResNet-50 is higher by 13.9% and the validation-loss rate lower by 0.151. This proves that the overall training effect of the Resnet-50 neural network is better than that of MobileNetV2. Basing on the near-infrared absorption spectroscopy, we also detect the protein content in hazelnuts, which represents an important parameter for evaluating their quality. A Kennard–Stone algorithm is used to classify a sample set. To elaborate a technique for the quantitative protein analysis of hazelnuts, we employ a partial least-squares method. The appropriate spectral data is preprocessed according to the methods of first derivative, second derivative and standard normal variate. The effect of these methods on the accuracy are compared. The results demonstrate that the model based on the first derivative is the best in case of the data referred to the overall spectral range. The correlation coefficients for the training and test sets are respectively equal to 0.938 and 0.965, whereas the root-mean-square errors for these sets amount respectively to 0.286 and 0.577. Our study testifies that the protein content in hazelnuts can be quickly and nondestructively detected using the near-infrared spectroscopy.

Keywords: neural networks, deep learning, partial least-squares method, near-infrared spectroscopy, nondestructive testing, protein content

UDC: 535-1+630*8+004.93

    1. Neus B, Cháfer M, Chiralt A and González-Martínez C, 2014. Hazelnut milk fermentation using probiotic Lactobacillus rhamnosus GG and inulin. Int. J. Food Sci. Technol. 49: 2553–2562. doi:10.1111/ijfs.12585
    2. Giovanna Canneddu, Luis Carlos Cunha Junior and Gustavo Henrique de Almeida Teixeira. 2016. Quality evaluation of shelled and unshelled macadamia nuts by means of near-infrared spectroscopy (NIR). J. Food Sci. 81: 1613–1621. doi:10.1111/1750-3841.13343
    3. Pannico A, Schouten R E, Basile B, Romano R, Woltering E J and Cirillo C, 2015. Non-destructive detection of flawed hazelnut kernels and lipid oxidation assessment using NIR spectroscopy. J. Food Engin. 160: 42–48. doi:10.1016/j.jfoodeng.2015.03.015
    4. Venkatachalam M and Sathe S K, 2006. Chemical composition of selected edible nut seeds. J. Agricult. Food Chem. 54: 4705–4714. doi:10.1021/jf0606959
    5. Delgado T, Malheiro R, Alberto Pereira J and Ramalhosa E, 2010. Hazelnut (Corylus avellana L.) kernels as a source of antioxidants and their potential in relation to other nuts. Indust. Crops Prod. 32: 621–626. doi:10.1016/j.indcrop.2010.07.019
    6. Ayvaz H, Temizkan R, Efe Genis H, Mortas M, Genis D O, Dogan M A, Nazlim B A, 2022. Rapid discrimination of Turkish commercial hazelnut (Corylus Avellana L.) varieties using near-infrared spectroscopy and chemometrics. Vibr. Spectrosc. 119: 103353. doi:10.1016/j.vibspec.2022.103353
    7. He Yizhi, Zhu Tiancheng, Wang Mingxuan and Lu Hanqing, 2021. On lemon defect identification with visual feature extraction and transfers learning. J. Data Analysis Inform. Process. 9: 233–248. doi:10.4236/jdaip.2021.94014
    8. Li Cong, Yu Guowei, Zhang Yuanjia and Ma Benxue, 2022. Research on recognition of stem/calyx and defects of dried Hami jujube based on ResNeXt and transfer learning. Food & Machinery. 38: 135–140.
    9. Liying Fan, Jun Ren, Yuting Yang and Limin Zhang, 2020. Comparative analysis on essential nutrient compositions of 23 wild hazelnuts (Corylus heterophylla) grown in Northeast China. J. Food Qual. 2020: 9475961. doi:10.1155/2020/9475961
    10. Levinson L R and Gilbride K A, 2011. Detection of melamine and cyanuric acid in vegetable protein products used in food production. J. Food Sci. 76: C568–C575. doi:10.1111/j.1750-3841.2011.02148.x
    11. Ji Xiao-fei, You Ming-hong, Bai Shi-qie, Li Da-xu, Lei Xiong, Wu Qi, Chen Li-min, Zhang Chang-bing, Yan Jia-jun, Yan Li-jun, Chen Li-li and Zhang Yu, 2019. Establishment of quantitative model for analyzing crude protein in phalaris arundinacea L. by near infrared spectroscopy (NIRS). Spectrosc. Spectr. Analysis. 39: 1731–1735. doi:10.1111/pai.13321/v3/response1
    12. Chu X L, Xu Y-P and Lu W-Z, 2008. Research and application progress of chemometrics methods in near infrared spectroscopic analysis. Chin. J. Analyt. Chem. 36: 702–709.
    13. Arndt M, Rurik M, Drees A, Bigdowski K, Kohlbache O and Fischer M, 2020. Comparison of different sample preparation techniques for NIR screening and their influence on the geographical origin determination of almonds (Prunus dulcis MILL). Food Control. 115: 107302. doi:10.1016/j.foodcont.2020.107302
    14. Ming-Zhi Zhu, Beibei Wen, Hao Wu, Juan Li, Haiyan Lin, Qin Li, Yinhua Li, Jianan Huang and Zhonghua Liu, 2019. The quality control of tea by near-infrared reflectance (NIR) spectroscopy and chemometrics. J. Spectrosc. 2019: 8129648. doi:10.1155/2019/8129648
    15. Eisenstecken D, Stürz B, Robatscher P, Lozano L, Zanella A and Oberhuber M, 2019. The potential of near infrared spectroscopy (NIRS) to trace apple origin: Study on different cultivars and orchard elevations. Postharvest Biol. Technol. 147: 123–131. doi:10.1016/j.postharvbio.2018.08.019
    16. Veronica Loewe M, Navarro R M, Cerrillo R M, Cerrillo N, Garcia-Olmo J, Garcia-Olmo J and Sánchez-Cuesta R, 2017. Discriminant analysis of Mediterranean pine nuts (Pinus pinea L.) from Chilean plantations by near infrared spectroscopy (NIRS). Food Control. 73: 634–643. doi:10.1016/j.foodcont.2016.09.012
    17. Chen H, Lin Z and Tan C, 2019. Classification of different animal fibers by near infrared spectroscopy and chemometric models. Microchem. J. 144: 489–494. doi:10.1016/j.microc.2018.10.011
    18. Swathi Sirisha Nallan Chakravartula, Chiara Cevoli, Balestra F, Fabbri A, Dalla Rosa M, 2019. Evaluation of drying of edible coating on bread using NIR spectroscopy. J. Food Engin. 240: 29–37. doi:10.1016/j.jfoodeng.2018.07.009
    19. Pérez-Marín D, Calero L, Fearn T, Torres I, Garrido-Varo A and Sánchez M-T, 2019. A system using in situ NIRS sensors for the detection of product failing to meet quality standards and the prediction of optimal postharvest shelf-life in the case of oranges kept in cold storage. Postharvest Biol. Technol. 147: 48–53. doi:10.1016/j.postharvbio.2018.09.009
    20. Moscetti R, Monarca D, Cecchini M, Ron P Haff, Contini M and Massantini R, 2014. Detection of mold-damaged chestnuts by near-infrared spectroscopy. Postharvest Biol. Technol. 93: 83–90. doi:10.1016/j.postharvbio.2014.02.009
    21. Moscetti R, Ron P Haff, Sirinnapa Saranwong, Monarca D, Cecchini M and Massantini R, 2014. Nondestructive detection of insect infested chestnuts based on NIR spectroscopy. Postharvest Biol. Technol. 87: 88–94. doi:10.1016/j.postharvbio.2013.08.010
    22. Jiaqi Hu, Xiaochen Ma, LinglingLiu, Yanwen Wu and Jie Ouyang, 2017. Rapid evaluation of the quality of chestnuts using near-infrared reflectance spectroscopy. Food Chem. 231: 141–147. doi:10.1016/j.foodchem.2017.03.127
    23. Williams P J and Kucheryavskiy S, 2016. Classification of maize kernels using NIR hyperspectral imaging. Food Chem. 209: 131–138. doi:10.1016/j.foodchem.2016.04.044
    24. Schönbrodt T, Mohl S, Winter G and Reich G, 2006. NIR spectroscopy-a non-destructive analytical tool for protein quantification within lipid implants. J. Control Release. 114: 261–267. doi:10.1016/j.jconrel.2006.05.022
    25. Cayuela-Sanchez J A and Palarea-Albaladejo J, 2019. Olive oil nutritional labeling by using Vis/NIR spectroscopy and compositional statistical methods. Innov. Food Sci. Emerg. Technol. 51: 139–147. doi:10.1016/j.ifset.2018.05.018
    26. Hacisalihoglu G, Freeman J, Armstrong P R, Seabourn B W, Porter L D, Settles A M and Gustin J L, 2020. Protein, weight, and oil prediction by single-seed near-infrared spectroscopy for selection of seed quality and yield traits in pea (Pisum sativum). J. Sci. Food Agricult. 100: 3488–3497. doi:10.1002/jsfa.10389
    27. Fan Y, Ma S and Wu T, 2020. Individual wheat kernels vigor assessment based on NIR spectroscopy coupled with machine learning methodologies. Infrar. Phys. Technol. 105: 103213. doi:10.1016/j.infrared.2020.103213
    28. Barbedo J G A, Guarienti E M and Tibola C S, 2018. Detection of sprout damage in wheat kernels using NIR hyperspectral imaging. Biosyst. Engin. 175: 124–132. doi:10.1016/j.biosystemseng.2018.09.012
    29. Zhang J, Dai L and Cheng F, 2019. Classification of frozen corn seeds using hyperspectral VIS/NIR reflectance imaging. Molecules. 24: 149. doi:10.3390/molecules24010149
    30. Moscetti R, Haff R P, Aernouts B, Saeys W, Monarca D, Cecchini M and Massantini R, 2013. Feasibility of Vis/NIR spectroscopy for detection of flaws in hazelnut kernels. J. Food Engin. 118: 1–7. doi:10.1016/j.jfoodeng.2013.03.037
    31. Tuncil Y E, 2020. Dietary fibre profiles of Turkish Tombul hazelnut (Corylus avellana L.) and hazelnut skin. Food Chem. 316: 126338. doi:10.1016/j.foodchem.2020.126338
    32. Navid Shakiba, Annika Gerdes, Nathali Holz, Soeren Wenck, René Bachmann, Tobias Schneider, Stephan Seifert, Markus Fischer, Thomas Hackl, 2022. Determination of the geographical origin of hazelnuts (Corylus avellana L.) by near-Infrared spectroscopy (NIR) and a low-level fusion with nuclear magnetic resonance (NMR). Microchem. J. 174: 107066. doi:10.1016/j.microc.2021.107066
    33. Jiaqi Hu, Xiaochen Ma, Lingling Liu, Yanwen Wu and Jie Ouyang, 2017. Rapid evaluation of the quality of chestnuts using near-infrared reflectance spectroscopy. Food Chem. 231: 141–147. doi:10.1016/j.foodchem.2017.03.127
    34. Moscetti R, Haff R P, Saranwong S, Monarca D, Cecchini M and Massantini R, 2014. Nondestructive detection of insect infested chestnuts based on NIR spectroscopy. Postharvest Biol. Technol. 87: 88–94. doi:10.1016/j.postharvbio.2013.08.010
    35. Zhou R B. Functional Principle and Technology of Plant Protein. Beijing: Chemical Industry Press, 2007.
    36. Yan Y L. Basis and Application of Near Infrared Spectroscopy. Beijing: China Light Industry Press, 2005.
    37. Fan W, Shan Y, Li G, Lv H, Li H and Liang Y, 2012. Application of competitive adaptive reweighted sampling method to determine effective wavelengths for prediction of total acid of vinegar. Food Analyt. Meth. 5: 585–590. doi:10.1007/s12161-011-9285-2
    38. Zhao Lihua, Gong Yuanyong, Zhang Jie, Lin Zhangbin, Wang Ying, Li Xuewu, Dai Xun and Jiang Yun, 2020. Ｒapid determination of quinoa seeds crude protein content using near infrared spectroscopy. Sci. Technol. Food Industry. 41: 233–236.
    39. Hongbo Li, Dapeng Jiang, Jun Cao and Dongyan Zhang, 2020. Near-infrared spectroscopy coupled chemometric algorithms for rapid origin identification and lipid content detection of Pinus Koraiensis seeds. Sensors. 20: 4905. doi:10.3390/s20174905

    Ядра фундука часто можуть бути неповними або деформованими та мати деякі інші дефекти. Ручні методи їхньої класифікації та розпізнавання або навіть застарілі підходи машинного навчання виявляють низьку ефективність розпізнавання та високий рівень неправильних оцінок. У цьому дослідженні ми застосували глибоке навчання до розпізнавання дефектів ядра фундука. Для навчання та розпізнавання використано дві згорткові нейромережеві системи MobileNetV2 і Resnet-50. Виявлено, що точність передбачення ResNet-50 для навчального набору покращена на 10,3%, а коефіцієнт втрат навчання зменшений на 0,041, порівняно з MobileNetV2. Крім того, точність перевірки, досягнута за допомогою ResNet-50, вища на 13,9%, а коефіцієнт втрат перевірки нижчий на 0,151. Це доводить, що загальний тренувальний ефект для нейронної мережі Resnet-50 кращий, ніж для MobileNetV2. За допомогою методу спектроскопії поглинання в близькому інфрачервоному діапазоні нами також вивчено вміст білка в фундуку, який є важливим параметром для оцінки його якості. Для класифікації вибірки використано алгоритм Кеннарда–Стоуна. Щоб розробити методику кількісного аналізу білка в фундуку, використано метод часткових найменших квадратів. Відповідні спектральні дані попередньо оброблено за методами першої похідної, другої похідної та стандартної нормальної змінної. Порівняно вплив цих методів на точність. Результати демонструють, що модель, заснована на першій похідній, є найкращою у разі даних, які стосуються всього спектрального діапазону. Коефіцієнти кореляції для навчальної та тестової вибірок дорівнюють відповідно 0,938 і 0,965, тоді як середньоквадратичні похибки для цих вибірок становлять відповідно 0,286 і 0,577. Наше дослідження засвідчує, що вміст білка в фундуку можна швидко виявити неруйнівним методом за допомогою близької інфрачервоної спектроскопії.

    Ключові слова: нейронні мережі, глибоке навчання, метод часткових найменших квадратів, спектроскопія близького інфрачервоного діапазону, неруйнівний контроль, вміст білка

Free download this article 

© Ukrainian Journal of Physical Optics ©