Multimodal fusion perception and posture regulation integration for intelligent detection of litchi fruit borer

This study presents a high-precision, non-destructive detection framework for identifying litchi fruit borer ( Conopomorpha sinensis ) through multimodal sensing and posture standardization. Visible/Near-infrared spectroscopy (VIS/NIRS), Hyperspectral imaging (HSI), and X-ray imaging were combined to capture complementary chemical, surface, and structural information. A dual-stage posture correction mechanism—comprising a passive V-shaped guide and a vision-based robotic arm—was developed to ensure stable fruit orientation across sensing modalities. Both data-level and feature-level fusion strategies were explored, and PLS-DA-based fusion models achieved 96.67% accuracy, clearly surpassing unimodal approaches. The proposed posture correction-multimodal fusion modeling framework improves detection robustness and offers strong potential for industrial-scale, non-destructive quality inspection of litchi and other irregularly shaped fruits.