VALIANT: A Vision‐Authenticity Language Framework Through Integrated Experts and Aligned Numerical‐Textual Descriptors for <i>Citri Reticulatae Pericarpium</i>

Citri reticulatae pericarpium (CRP; Chenpi), the dried peel of Citrus reticulata , is a traditional Chinese medicine whose value increases with storage age. This price premium incentivizes fraudulent aging through cosmetic treatments, high‐temperature fermentation, tea infusions, or chemical dyes. Expert sensory appraisal is subjective (approximately 83% accuracy), while vision‐only deep models—effective in controlled settings (90%–98% accuracy)—often degrade under real‐world smartphone conditions and against sophisticated counterfeits. We present VALIANT, a multimodal vision‐language framework for smartphone images that enables age classification (1, 6, 10, 15, and 20 years) and authenticity verification against standard operating procedure (SOP) mimics. VALIANT integrates three modalities: (1) visual embeddings from a dual‐expert system (Florence‐2 for global context, and Swin Transformer for local detail), (2) numerical fingerprints, including CIE 1976 L * a * b * (CIELAB) color statistics, gray‐level co‐occurrence matrix (GLCM) texture properties, and Laplacian variance, and (3) textual attributes generated by GPT‐4o‐mini. A gated cross‐attention module with contrastive alignment enhances semantic coherence and robustness to imaging noise. Evaluated on 942 smartphone images (761 genuine, 181 SOP mimics), VALIANT achieves 88.7% accuracy for genuine samples and 92.2% for mixed data, representing 4%–5% significant improvements over vision‐transformer baselines. This framework enables portable, rapid, and point‐of‐purchase verification to improve market integrity and supplier credibility.