Improved CycleGAN-Based Data Augmentation for Loquat Disease Classification

With the advancement of agricultural intelligence, deep learning-based plant disease recognition technology has increasingly become a crucial approach for managing crop health. However, due to challenges associated with data collection and the high cost of annotation, the available dataset for loquat disease classification is limited in scale and diversity, thereby restricting the effectiveness of deep learning model training. To address these challenges, this study proposes an improved cycle-consistent adversarial network model integrated with manual data augmentation techniques to enhance both the quality and diversity of the dataset. Experimental results demonstrate that the proposed data augmentation method significantly enhances the convergence speed, generalization ability, stability, and class balance of classification models. Furthermore, appropriate data augmentation can substantially improve the classification performance of small-sample datasets.