Advanced AI-driven detection of cinnamon powder adulteration using near-infrared spectroscopy and deep learning image recognition technique

Cinnamon powder is a crucial spice widely used in various food products. Due to the complexity of its origin, production, and processing, particularly for Cinnamomum verum , it commands a relatively high market price. To reduce costs, some manufacturers may adulterate cinnamon powder with nut-based powders. This practice constitutes not only food fraud but also poses allergen risks. To address this issue, this study developed the NIR-PP model, which is based on near-infrared (NIR) spectroscopy combined with chemometric techniques, including principal component analysis (PCA) and partial least squares regression (PLSR). In addition, the NIR-DL model, which utilizes deep learning-based multivariate classification methods for NIR spectral analysis, and the Image-DL model, which is based on deep learning for image recognition, were also established to analyze cinnamon powder samples. This study analyzed 75 samples, generating 75 NIR spectral data points and 475 image data points, and evaluated adulteration using three models. The results showed that the NIR-PP approach achieved an R 2 value of 0.99 for detecting adulterants. The NIR-DL model achieved an average F1-score of 0.978, whereas the Image-DL model achieved an average F1-score of 0.972. This study not only demonstrated the exceptional accuracy of the NIR-PP in chemometric analysis but also highlighted the effective application of deep learning techniques for cinnamon powder adulteration detection. It provides two reference models, NIR-DL and Image-DL, suitable for NIR data analysis and image analysis, respectively. These models significantly enhance real-time detection, accuracy, and convenience, while addressing the limitations of existing models. Furthermore, they exhibit excellent scalability and potential for broader applications. • Some producers adulterate cinnamon with hazelnut and walnut shell powder. • These powders are nut allergens, one of eight major types, and pose safety risks. • This study compares NIR-PP, NIR-DL, and Image-DL for detecting food adulteration. • Image-DL uses imaging tech to rapidly detect adulterants and simplify detection. • Deep learning helps detect fraud, reduce allergen risks, and improve food safety.