Document Type : Research Paper
Authors
Department of Biosystems Engineering, Faculty of Agriculture and Enviromental Resources, University of Mohaghegh Ardabili, Ardabil, Iran.
Abstract
Keywords
Main Subjects
EXTENDED ABSTRACT
The primary objective of this research was to develop and evaluate a rapid, non-destructive method for predicting the soluble solids content (SSC) in two distinct plum cultivars, namely Khormaei and Khoni, using visible and near-infrared (VIS/NIR) spectroscopy. The study aimed to overcome the limitations of traditional destructive methods, such as the refractometer, which are time-consuming and unsuitable for real-time quality control in the post-harvest industry. Furthermore, a key goal was to enhance the predictive performance of models by employing a novel combination of advanced meta-heuristic algorithms and machine learning techniques for optimal wavelength selection and model development.
A total of 80 plum samples, consisting of 40 samples from each of the Khormaei and Khoni cultivars, were utilized for this study. The fruit samples were subjected to VIS/NIR spectroscopy in the range of 350 to 1100 nm to capture their absorption spectra. The corresponding SSC values for each sample were simultaneously measured using a digital refractometer as the reference method. To address the challenge of high dimensionality and multicollinearity inherent in spectral data, a comprehensive feature selection strategy was implemented. A suite of five meta-heuristic algorithms, including Ant Colony Optimization (ACO), Particle Swarm Optimization (PSO), Genetic Algorithm (GA), League Championship Algorithm (LCA), and Imperialist Competitive Algorithm (ICA), was employed in conjunction with a Support Vector Machine (SVM) model. The LCA algorithm was found to be the most effective in identifying a subset of key wavelengths that were most influential in predicting SSC. For model development and comparison, both Partial Least Squares Regression (PLSR) and Artificial Neural Network (ANN) models were employed. To ensure the robustness and generalizability of the models, data splitting was performed 200 times with a ratio of 60% for training, 20% for validation, and 20% for testing. Additionally, the impact of more than 12 different pre-processing techniques, such as Moving Average, Savitzky-Golay filtering, and Standard Normal Variate (SNV), on model accuracy was systematically investigated. The parameters for the ANN model, including the number of hidden layers and neurons, were optimized through a systematic process to prevent overfitting and achieve the best performance.
The results demonstrated that the non-destructive VIS/NIR method combined with chemometric modeling is highly effective for predicting SSC in both plum cultivars. The performance of the models varied depending on the cultivar, the pre-processing method, and the model type. The ANN model consistently outperformed the PLSR model, especially when applied to the reduced spectral data set. The highest prediction accuracy was achieved by the ANN model using the key wavelengths selected by the LCA algorithm, with the Moving Average pre-processing method. For the Khormaei cultivar, this model yielded a coefficient of determination of R2=0.989 and a ratio of performance to deviation of RPD=6.08, indicating excellent predictive capability. Similarly, for the Khoni cultivar, the model’s performance was also highly satisfactory. The selection of optimal wavelengths by the meta-heuristic algorithms significantly improved model performance while reducing computational complexity. The LCA-SVM combination proved to be particularly powerful in isolating the most relevant spectral information for SSC prediction.
The study successfully demonstrated that VIS/NIR spectroscopy coupled with advanced meta-heuristic feature selection and ANN modeling is a highly accurate and reliable method for the non-destructive determination of SSC in plum cultivars. The combination of the LCA algorithm for wavelength selection with an ANN model offers a novel and powerful approach for creating robust and efficient predictive models. This methodology not only improves prediction accuracy but also reduces the computational requirements, making it a viable and practical tool for real-time quality control applications in the post-harvest handling and food processing industries. The findings confirm that this non-destructive technique is a promising alternative to traditional destructive methods and can be effectively implemented to ensure product quality and consistency.
Conceptualization, M.L.-A. and Y.A.-G.; methodology, M.L.-A.; software, M.L.-A.; validation, Y.A.-G.; formal analysis, M.L.-A. and Y.A.-G.; investigation, M.L.-A.; resources, M.L.-A.; data curation, M.L.-A. and Y.A.-G.; writing—original draft preparation, M.L.-A. and Y.A.-G..; writing—review and editing, Y.A.-G.; visualization, M.L.-A.; supervision, Y.A.-G.; project administration, Y.A.-G.; funding acquisition, Y.A.-G.All authors have read and agreed to the published version of the manuscript.
All authors contributed equally to the conceptualization of the article and writing of the original and subsequent drafts.
This research was supported by the University of Mohaghegh Ardabili, which is greatly appreciated.
Not applicable.
The authors declare that they have no known competing financial interest or personal relationship that could have appeared to influence the work reported in this paper.
)