Nondestructive Discrimination of Pepper Seed Varieties Using Fluorescence Hyperspectral Imaging with Detrending—Competitive Adaptive Reweighted Sampling (DT-CARS) and Dung Beetle Optimization—Support Vector Machine (DBO-SVM)

Efficient identification of pepper seed varieties is not only a crucial approach to enhancing the value of the pepper industry but also a practical requirement for safeguarding the legitimate rights and interests of seed producers. This study aims to investigate the feasibility of using fluorescence hyperspectral imaging for the identification of pepper seed varieties. Firstly, fluorescence hyperspectral images of pepper seeds from five varieties were collected. The entire seed image was designated as the region of interest, and its average spectrum was extracted as the fluorescence spectrum for that individual seed. During data processing, the results of various preprocessing methods, such as Savitzky–Golay smoothing, multiplicative scatter correction, and standard normal variate transformation, were compared and analyzed. Concurrently, the dimensionality reduction effects of algorithms such as successive projections algorithm, competitive adaptive reweighted sampling, interval variable iterative space shrinkage approach, and random frog, along with interval combination optimization, were thoroughly investigated. Finally, intelligent optimization algorithms, including Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), and Dung Beetle Optimizer (DBO), were introduced to select the optimal hyperparameters for the Support Vector Machine (SVM), thereby constructing a classification framework for nondestructive detection. Experimental results indicated that the DBO-SVM model, established based on spectral features processed and extracted via detrending—competitive adaptive reweighted sampling (DT-CARS), exhibited the best model performance, achieving a mean test-set accuracy of 99.8 ± 0.4% across 10 repeated optimization runs. These findings demonstrate the feasibility of fluorescence HSI technology for discriminating among pepper seed commercial lots as a proof-of-concept.