Geographical authentication of medicinal plant materials remains challenging because multispectral variables are often highly collinear and sample grouping can complicate reliable model validation. Existing correlation-based feature-selection strategies also require careful adaptation to multiclass problems to avoid artificial ordering of class labels and information leakage during model development. Therefore, this study aimed to develop a compact and leakage-controlled multispectral learning framework for geographical-origin discrimination. This study analyzed 800 physical Pinellia ternata samples from Gansu Xihe, Sichuan Neijiang, Sichuan Chengdu, and Chongqing Dianjiang (200 samples per origin). Each physical sample was represented by 31 mean grayscale intensities calculated from Otsu-segmented multispectral regions of interest. A Pearson-correlation-guided deep multilayer perceptron (PCG-DeepMLP) was constructed by estimating one-vs-rest band relevance and inter-band redundancy only within the training data. The key methodological innovation is a unified multiclass-aware, relevance–redundancy spectral-learning framework in which class-specific one-vs-rest Pearson relevance is coupled with inter-band redundancy control and embedded within leakage-controlled grouped model development. By learning the spectral subset exclusively from each training partition before nonlinear classification, the framework produces compact and complementary multispectral signatures while preserving multiclass structure and strict independence of held-out groups. Model and feature-selection settings were chosen by three-fold grouped cross-validation within each training partition. PCG-DeepMLP retained 9–21 bands and achieved the highest mean accuracy (0.9812 ± 0.0135), macro-F1 (0.9812 ± 0.0135), Matthews correlation coefficient (MCC; 0.9752 ± 0.0179), and macro-AUC (0.9994 ± 0.0006) among seven models. Its macro-F1 was higher than that of 1D-CNN, 1D-ResNet, full-band MLP, PLS-DA, and random forest after Holm correction. Performance was estimated through a strict nested group-wise internal validation scheme, with every outer test fold remaining isolated from feature selection, preprocessing, and model optimization. These findings demonstrate that multiclass-aware relevance–redundancy learning can retain complementary Pinellia ternata origin-discriminative information in a compact and stable spectral representation, enabling accurate geographical-origin authentication while providing a principled basis for reduced-channel acquisition and future independent multi-batch validation.

Multispectral imaging systems often exhibit wavelength-dependent focus responses because of chromatic aberration, nonuniform spectral sensitivity, and scene-dependent reflectance, making single-band autofocus inadequate for optimizing the complete spectral image cube. This study proposes a nonlinear band-correlation guided deep reinforcement learning framework for autofocus optimization in a self-developed 31-band multispectral imaging system covering 360–980 nm. Through-focus response curves are first extracted from all spectral channels, and a nonlinear inter-band dependency matrix is constructed to characterize complementary and redundant focusing information across wavelengths. A compact subset of representative bands is then selected and embedded into the reinforcement-learning state to guide closed-loop motor control. The agent jointly determines focusing direction, displacement, and stopping time while maximizing a global multispectral focus objective that considers mean sharpness, inter-band consistency, and worst-band degradation. During online autofocus, only the representative bands are acquired at intermediate positions, whereas the complete 31-band cube is captured after convergence. The framework is evaluated using standardized optical targets and spectrally heterogeneous scenes through focal-position error, MTF, full-band focus retention, acquisition time, and robustness. The proposed design provides an efficient computational-photonics solution for broadband multispectral autofocus with reduced band-acquisition and motor-search overhead.