Research on millet origin identification model based on improved parrot optimizer optimized regularized extreme learning machine

Abstract

To achieve nondestructive identification of millet origins, near-infrared spectroscopy technology was employed to obtain the original spectral data of millet. By integrating the Parrot Optimizer (PO) with the Regularized Extreme Learning Machine (RELM), the model achieved an accuracy of 83.67 % in millet origin identification. To further enhance model performance, this study incorporated strategies such as chaotic mapping and adaptivity into PO, resulting in the Improved Parrot Optimizer (IPO). The IPO was then combined with RELM to construct the IPO-RELM model, which significantly improved the model's generalization capability and robustness. Experimental results demonstrated that the IPO-RELM model outperformed the RELM model, achieving an accuracy of 98.33 %, a precision of 98.49 %, a recall of 98.33 %, an F1 score of 98.41 %, and a Kappa coefficient of 98 %, representing respective improvements of 11.32 %, 7.92 %, 11.32 %, 9.62 %, and 13.90 % over the traditional RELM model. Furthermore, the performance of the IPO-RELM model was validated using two publicly available datasets, confirming its superiority over the conventional RELM model. Compared to the PO algorithm, the IPO algorithm exhibited enhanced global search and local optimization capabilities with faster convergence speed. The IPO-RELM model accurately and efficiently identified millet origin information, providing robust support for ensuring millet quality and safety, while also contributing to the protection of the uniqueness and market value of geographically indicated agricultural products.