Enhancing Mango Leaf Disease Classification: ViT, BiT, and CNN-Based Models Evaluated on CycleGAN-Augmented Data

Mango farming, crucial to agriculture in Bangladesh, faces severe threats from leaf diseases, notably anthracnose and powdery mildew. For diseases to be effectively managed, it must be accurately identified. This work uses a novel approach called CycleGAN-generated data augmentation to classify these diseases. Exceptional results were achieved using various models, including ViT-B/16, ViT-B/32, BiT-M-R50x1, BiT-s-R50x1, VGG16, ResNet152, ResNet50, InceptionV3, and ResNet50V2, although only ViT-B/16, ViT-B/32, and BiT-M-R50x1 reached 100% accuracy. This represents a significant development in the detection of mango leaf diseases and demonstrates CycleGAN’s potential to improve model performance. The acclaimed image classification Vision Transformer models ViT-B/16 and ViT-B/32 performed exceptionally well at differentiating between healthy and damaged leaves. This performance was echoed by the adaptable BiT-M-R50x1 model, demonstrating the potency of our strategy. These findings hold great promise for Bangladesh’s agriculture, potentially revolutionizing mango leaf disease management, and addressing a significant concern for the nation’s mango industry. Future research may expand this methodology to encompass more mango leaf diseases, further enhancing its relevance. This study underscores the transformative potential of advanced deep learning techniques in safeguarding agricultural yields and ensuring food security in mango-producing regions of Bangladesh.