Abstract: Slight mechanical damage is an important factor affecting the quality of ginger. In certain situations, such damage may be hidden beneath the outer layer of the skin, which is often difficult to identify directly by routine examination. A non-destructive detection method for ginger damage based on near-infrared (NIR) spectroscopy was proposed in this study. For this purpose, four cultivars of ginger with artificial mechanical damage were selected. First, different spectral preprocessing methods including standard normal variate (SNV), multiple scattering correction, baseline correction, and Savitzky-Golay smoothing (SGS) were compared, and then classification models were established by three machine learning algorithms: support vector machine (SVM), K-nearest neighbor, and random forest. The results showed that the combination of SGS and SNV greatly improved the recognition accuracy of multiple models, demonstrating strong performance advantages. The SGS + SNV-SVM model was the most effective in detecting ginger damage, with an average recognition accuracy of over 92% for all four ginger cultivars (100% for two cultivars). This model provides a technical reference for the accurate identification of damaged ginger.

Key words: near-infrared spectroscopy; model identification; ginger; damage; spectral preprocessing methods