Detection of bruised loquats based on reflectance, absorbance and Kubelka–Munk spectra

• Published in: Journal of food measurement & characterization Vol. 17; no. 2; pp. 1562 - 1575
• Main Authors: Li, Bin, Han, Zhaoyang, Wang, Qiu, Yang, Akun, Liu, Yande
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.04.2023; Springer Nature B.V
• Subjects: Absorbance; Adaptive sampling; Algorithms; Artificial neural networks; Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Classification; Decay; Economic impact; Engineering; Food quality; Food safety; Food Science; Genetic algorithms; Hyperspectral imaging; K-nearest neighbors algorithm; Machine learning; Original Paper; Reflectance; Spectra; Support vector machines; Transportation; Absorbance; Loquats; Bruised; Kubelka–Munk; Hyperspectral imaging; Reflectance
• ISSN: 2193-4126; 2193-4134
• DOI: 10.1007/s11694-022-01717-3

Bruise is one of the main problems in the classification and processing of loquats after harvest, causing decay and deterioration of normal loquats during transportation. It reduces the economic value of the loquats and raises food quality and safety problems. Therefore, improving the identification rate of bruised loquats can effectively reduce the economic loss caused by transportation and storage. In this study, hyperspectral imaging was used to collect reflectance (R), absorbance (A), and Kubelka–Munk (KM) spectra of bruised loquats for bruised grade detection. “Gaussian” filter (GF), baseline offset correction (BOC), maximum normalization (MAN), Savitzky-Golay (SG), and multiplicative scatter correction (MSC) were used to preprocess the original spectral data. Competitive adaptive reweighted sampling (CARS), genetic algorithm (GA), successive projections algorithm (SPA), and uninformative variables elimination (UVE) were used to reduce the dimension of spectral data to obtain the characteristic wavelength. Random forest (RF), extreme learning machine (ELM), least square support vector machines (LS-SVM), and k-nearest neighbor (KNN) algorithm were used to establish the classification model of bruised grade in loquats. By optimizing the model based on all model classification results, the best model (MIX) is obtained. By comparison, the results revealed that the MIX model showed the lowest errors, better stability and generalization ability, with an accuracy of 100%. Consequently, it also provides a theoretical reference for the rapid, nondestructive, and high-precision fruit online detection technology in the future.