A field-acquired annotated dataset for robust and generalizable pomegranate disease detection using generative learning

Pomegranates are popular in India for their health benefits and high levels of antioxidants and vitamins. Detecting diseases in pomegranates is important for maintaining their quality. However, inconsistent dataset annotations and challenging field conditions make it difficult to collect high-quality images, limiting deep learning model performance. To address this, we introduce a Dual GAN–Diffusion-based generative learning system that creates high-quality, diverse images to improve pomegranate disease datasets. We collect and process real-time data using a GAN to generate diverse and realistic images for detecting and diagnosing pomegranate diseases. With a Fr´echet Inception Distance (FID) of 22.40, Inception Score (IS) of 49.75, LPIPS Diversity of 69.86%, and classification Accuracy of 76%, the perform The system achieved a Fr´echet Inception Distance (FID) of 22.40, Inception Score (IS) of 49.75, LPIPS Diversity of 69.86%, and classification accuracy of 76%. These results show clear improvements over baseline GAN models. This will show that our research approach can generate diverse, realistic samples, improve the system, and make healthy and diseased classification more scalable and reliable in real-world agricultural settings.