A Novel Approach for Quantifying the Wrinkling Degree of Cured Tobacco Using Unsupervised Depth Estimation

To address the issue of position differentiation in the digital transformation of the tobacco industry based on the wrinkling degree of cured tobacco, this study introduces the concept and definition of the “wrinkling degree of cured tobacco”. The determination parameters for the wrinkling degree are identified as the arithmetic mean deviation and the maximum difference of surface pixels. An enhanced unsupervised depth estimation model is developed to measure the wrinkling degree of cured tobacco, and its effectiveness is validated through the accuracy of a test set containing known tobacco leaf positions. The experimental results demonstrate that the proposed method achieves an accuracy of 89.83% when using the wrinkling degree of cured tobacco as a reference for location differentiation, surpassing other networks by 11.18% in terms of accuracy. The findings indicate that the unsupervised depth estimation method effectively measures the wrinkling degree of cured tobacco, thereby addressing the challenge of distinguishing leaf positions in the digital transformation process.