Enhancing tobacco leaf similarity research through multi-modal feature fusion

The study proposed a method for identifying tobacco leaves with similar quality to stored leaves. It calculated leaf similarity using multi-modal fusion. It extracted near-infrared spectral features with a 1D-CNN network and image features with the correlation function in OpenCV. Then, it applied the attention mechanism feature fusion method to combine these features. The method calculated the similarity between tobacco leaves using Euclidean distance, which allows for the identification of replacement leaves most similar to the target leaves. This approach effectively fulfills the objective of sustaining the cigarette leaf group formula during maintenance by addressing the requirement to identify comparable tobacco leaves in instances where a specific raw material is lacking.