PotatoGANs: Utilizing Generative Adversarial Networks, Instance Segmentation, and Explainable AI for Enhanced Potato Disease Identification and Classification

We introduce PotatoGANs, a hybrid augmentation approach using CycleGAN and Pix2Pix to generate synthetic diseased potato images from healthy samples, enhancing dataset diversity and model generalization while reducing data collection costs. To support model interpretability, we combine GradCAM, GradCAM++, and ScoreCAM with DenseNet169, ResNet152 V2, and InceptionResNet V2, offering transparent visual explanations of model predictions. Unlike existing work focused solely on leaf-level analysis, our method addresses whole-crop disease localization using advanced segmentation tools like Detectron2. Validated by the Bangladesh Agricultural Research Institute, this study aims to support the advancement of agricultural disease diagnosis and management in Bangladesh.