Early detection of mold in cured tobacco leaves based on temporal hyperspectral imaging

The early detection and warning of mold in tobacco leaves are critical for minimizing losses caused by mold. Existing studies primarily focus on spectral feature analysis at a single time point, overlooking the dynamic evolution of the mold process. To address this limitation, we propose a novel early detection method for mold in tobacco leaves using temporal hyperspectral imaging. First, hyperspectral data of moldy tobacco leaf samples were collected at different time points. Spectral correction and image alignment methods were applied to enhance data quality and ensure spatial consistency across hyperspectral images acquired at different times. To further capture the dynamic characteristics of mold, cumulative energy, and backward short-term energy features are introduced and combined with first- and second-order derivatives, enabling a comprehensive depiction of the temporal behavior of spectral reflectance while effectively identifying local anomalies and long-term trends during the early stages of mold. Additionally, a conditional Wasserstein generative adversarial network with gradient penalty (CWGAN-GP) addresses data imbalance issues, incorporating gradient penalties and conditional information to enhance the quality of generated samples significantly. Experimental results demonstrate that the proposed method effectively detects spectral changes associated with the early stages of mold in tobacco leaves, offering a promising approach for mold warning and real-time monitoring. This study enriches the theoretical framework of hyperspectral image analysis and provides valuable technical support for tobacco quality control.