Canker is a common disease of navel oranges that is visible before harvest, and penicilliosis is a common disease occurring after harvest and storage. In this research, the typical fruit surface, canker spots, penicillium spore, and hypha of navel oranges were, respectively, identified by hyperspectral imaging. First, the light intensity on the edge of samples in hyperspectral images was improved by spherical correction. Then, independent component images and weight coefficients were obtained using independent component analysis. This approach, combined with use of a genetic algorithm, was used to select six characteristic wavelengths. The method achieved dimension reduction of hyperspectral data, and the testing time was reduced from 46.21 to 1.26 s for a self-developed online detection system. Finally, a deep learning neural network model was established, and the four kinds of surface pixels were identified accurately.
Performance of portable near-infrared spectrometers is easily affected by various factors such as on-site environment, which results in a certain deviation in the on-site predicted results. Partially least square regression (PLSR), as a multiple linear regression method, has been widely used for the analysis of near-infrared (NIR) spectroscopy. However, due to the nonlinear characteristics of the relationship between spectral data and dependent variables, PLSR can easily lead to model errors. Stability and predictability decreased when PLSR is applied in on-site quick detection. How to reduce the errors caused by various environmental factors in the use of portable near-infrared spectrometers is a key issue in promoting the wide application of rapid detection technology based on near-infrared spectral analysis. In this study, the absorption spectra data of glucose solutions of different concentrations are collected by a portable near-infrared spectrometer. Several nonlinear correction algorithms are applied to study the effect of environmental interference during the measurement process. Firstly, the collected spectra data is preprocessed. Secondly, the data is modeled by nonlinear correction algorithms such as optimized artificial neural network (ANN), support vector regression (SVR), and random forest (RF). The impact of different models is compared with to the results using PLSR. It is found that compared with the PLSR linear method, the ANN, SVR and RF nonlinear correction algorithm can eliminate the interference of environmental factors in different degrees. Therefore, ANN, SVR and RF algorithm can improve the prediction accuracy of the model. This study shows that the use of nonlinear correction algorithm for data modeling of portable near-infrared spectrometers can effectively improve the predictive performance of the model.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.