Study on Recognition of Typical Curing Stages of Jiangxi Tobacco Leaves Based on Image Processing

This study aims to explore the visual characteristic changes of tobacco leaf images in the typical drying stages dataset of Jiangxi tobacco leaves. By establishing a recognition model for tobacco leaf drying stages, it achieves accurate identification of the drying stages and enhances the core technology of intelligent tobacco leaf drying in Jiangxi, promoting the high-quality development of tobacco leaf drying. Firstly, the color features and texture features of the tobacco leaf image dataset are extracted using MATLAB 2016. Then, feature variable clustering and correlation analysis are carried out using SAS 9.4 to select the optimal features. Finally, the selected features are input into classification models including Extreme Learning Machine (ELM), Genetic Algorithm-based Support Vector Machine (GA-SVM), Particle Swarm Optimization-based Backpropagation Neural Network (PSO-BP), and deep learning model AlexNet for comparative analysis. Using the selected color features 2G-R-B and G, and the texture features gradient mean and gradient distribution non-uniformity as inputs to ELM, GA-SVM, and PSO-BP classification models, the accuracy rates are 83.00%, 85.63%, and 90.92% respectively, while the deep learning model AlexNet achieves an accuracy rate of 90.67%.The ELM, GA-SVM, PSO-BP, and AlexNet models constructed based on the Jiangxi tobacco leaf image dataset can all achieve the classification of the typical drying stages of Yunyan 87 tobacco leaves, with AlexNet performing the best. This research lays the foundation for further in-depth study of machine learning models, and has significant implications for accurate judgment of tobacco leaf status during the drying stages.