نوع مقاله : مقاله پژوهشی
نویسندگان
گروه مهندسی بیوسیستم، دانشکده کشاورزی و منابع طبیعی، دانشگاه محقق اردبیلی، اردبیل، ایران.
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
This study aimed to develop and evaluate a non-destructive approach for predicting peroxidase (POD) enzyme activity in two plum cultivars—‘Khormaei’ and ‘Khouni’—using Vis/NIR spectroscopy combined with machine learning techniques. After acquiring absorbance spectra, the data were modeled using Partial Least Squares Regression (PLSR) and Support Vector Machine (SVM) algorithms across both the full spectral range and reduced dimensions. Model performance was assessed based on the coefficient of determination (R²), root mean square error (RMSE), and the ratio of performance to deviation (RPD). To enhance model efficiency and reduce data dimensionality, SVM was integrated with metaheuristic algorithms for effective wavelength selection. Among these, Particle Swarm Optimization (PSO) emerged as the most effective algorithm. The results demonstrated that dimensionality reduction strategies, by eliminating noise and redundant information, significantly improved the predictive accuracy of the models. Findings indicated that the optimal model varied between cultivars. For the ‘Khormaei’ cultivar, the SVM-R model with an RBF kernel and median filter preprocessing yielded the best performance, achieving an RPD of 2.687, categorizing it as an excellent model and underscoring the relevance of nonlinear approaches for this cultivar. Conversely, for the ‘Khouni’ cultivar, the best result was obtained using the linear PLSR model with normalization preprocessing, achieving an RPD of 2.29, which is considered very good. This contrast highlights that optimal model selection is inherently dependent on the intrinsic characteristics of each product. Ultimately, the study confirms that Vis/NIR spectroscopy is a powerful tool for rapid and non-destructive postharvest quality monitoring of agricultural products.
کلیدواژهها [English]
)