A Quality Identification Method for Chinese Herbal Slices Integrating Fuzzy Reasoning–Based Open‐Set Recognition and Support Vector Machine

In recent years, quality evaluation of Chinese Herbal Slices (CHS) has primarily relied on morphological traits, chemical composition analysis, and bioactivity assessments, with nondestructive techniques, such as near‐infrared (NIR) spectroscopy and X‐ray imaging, remaining underutilized. Meanwhile, open‐set recognition (OSR), a research hotspot in pattern recognition over the past decade, enables models to not only classify test samples from known classes in the training set but also effectively reject unknown classes. Here, collected CHS classes are defined as known classes, while uncollected classes are treated as unknown classes excluded from the training set. Simultaneously, quality assessment of known CHS classes is required. Existing OSR algorithms, predominantly based on support vector machine (SVM) and deep learning frameworks, are mainly applied to image recognition, with limited exploration in spectral analysis. This study proposes a novel method integrating a fuzzy inference–based OSR classifier with SVM for CHS quality evaluation. First, NIR spectra of CHS were acquired using a FLAMENIR spectrometer. Spectral noise was reduced via Gaussian smoothing, followed by dimensionality reduction to 5 features using principal component analysis (PCA). Subsequently, a hierarchical classifier was constructed: The OSR classifier distinguishes known and unknown classes, while SVM evaluates the quality of known‐class CHS. Across four scenarios on the CHS dataset containing 1700 samples, the proposed method outperforms OSR algorithms and state‐of‐the‐art 1D deep learning algorithms in terms of overall recognition accuracy (ORA), F1, and kappa. These algorithms include weighted‐SVDD, open‐set version of nearest neighbor classifier (OSNN), metric learning, fuzzy reasoning (FR), 1D‐CNN, and 1D‐ResNet integrated with the fusion of collective and extreme information (FCEI).