Abstract: To address the problems of low recognition accuracy, sparse image features, and low efficiency of manual discrimination during the tobacco leaf curing process, a dedicated dataset containing images of the entire tobacco leaf curing process was established, and the curing process was divided into 10 curing stages according to local curing protocols. To address the limitations of a single model in feature representation, ConvNeXt, RegNet, and Swin Transformer were selected as the base models for ensemble learning, Support Vector Machine(SVM) was used as the meta-model for ensemble learning, and the Differential Evolution(DE)algorithm was introduced to automatically optimize the hyperparameters of the SVM meta-model, thereby constructing an ensemble learning-based tobacco leaf curing stage recognition model, namely CT-DE-SVM (CNN-Transformer-DE-SVM). The results showed that: 1)In the ablation experiments, the CT-DE-SVM ensemble model achieved an accuracy of 95.12% and a macro F1-score of 0.9515, indicating that the stacking ensemble strategy can achieve multi-dimensional feature complementarity by fusing structurally diverse deep learning models; 2)The precision of the CT-DE-SVM ensemble model for all 10 curing stages exceeded 91.00%, and the F1-scores were all above 0.90, indicating that the model possesses strong feature discrimination capability across different curing stages; 3)On the test set, compared with single recognition models such as ResNet50, the CT-DE-SVM ensemble model performed best across all metrics, with accuracy of 95.12%, macro precision of 95.67%, macro recall of 94.84%, and macro F1-score of 0.951 5, respectively. The proposed CT-DE-SVM ensemble learning model can effectively identify tobacco leaf curing stages, providing technical support for intelligent tobacco curing platforms.