Deep Learning-Based Classification of Tobacco Leaf Alcoholization Levels Resistant to Lighting Interference

The alcoholization level of tobacco leaves is a critical parameter in the tobacco processing industry. Currently, the grading of tobacco alcoholization levels relies primarily on manual assessment, which is time-consuming, labor-intensive, and often lacks accuracy. This paper introduces a tobacco leaf alcoholization grading method based on computer vision and deep learning, aiming to achieve automatic alcoholization level determination while being resilient to lighting interference. By combining an attention mechanism with an enhanced FC
⁴
(Fully Convolutional Color Constancy with Confidence-weighted Pooling) model, the consistency of tobacco leaf color is maintained under varying lighting conditions. Additionally, the performance of the Back Progagation Neural Network is optimized using the particle swarm optimization to accurately map the color and texture features of the images to the classification results. Using tobacco leaves of known grades from the repository as experimental subjects, the test results indicate that the method proposed exhibits a higher classification accuracy compared to existing approaches.