An efficient nondestructive detection method of rapeseed varieties based on hyperspectral imaging technology

In response to the diverse requirements for rapeseed varieties in different fields and the prevalence of counterfeit seeds, efficient nondestructive detection methods are essential. Hyperspectral imaging (HSI) is widely used for this purpose, but its high dimensionality and redundant information complicate practical applications. This study proposes a dimensionality reduction algorithm that first selects feature wavelength intervals and then extracts features. The modified interval random frog (miRF) conducts supervised training on labeled spectral data to evaluate and select important wavelength intervals, capturing interactions between features while eliminating redundancy. Additionally, kernel principal component analysis (KPCA) addresses the nonlinear relationships among the selected intervals by mapping the data into a high-dimensional space, revealing its intrinsic structure and enhancing model generalization. This integrated approach constructs an optimized, streamlined feature space, improving detection capabilities for rapeseed varieties. The dimensionality reduction results of KPCAmiRF are also analyzed, and a strategy of feature selection followed by extraction is established. Furthermore, nature-inspired optimization algorithms, including hippopotamus optimization (HO), goose optimization (GOOSE), and artificial gorilla troop optimization (GTO), are introduced to refine hyperparameter selection and create a robust framework for efficient nondestructive detection. Ultimately, the GOOSE-SVC model established based on the spectral features extracted by miRF-KPCA demonstrated superior model performance and achieved an accuracy rate of 98.96% on the prediction set. The results validate the potential of HSI in rapeseed variety detection and present an innovative method for dimensionality reduction of hyperspectral data.