In the automation of agricultural disease segmentation, deep learning techniques play a crucial role. However, these applications often face overfitting when dealing with new conditions, leading to decreased segmentation performance. In potato farming, where diseases significantly impact yields, it is essential to identify these diseases quickly and accurately for the agricultural economy.
To address the limitations of traditional data augmentation methods like rotation, flipping, and translation, we propose a novel approach named PotatoGANs. This method employs two types of Generative Adversarial Networks (GANs) to generate synthetic potato disease images from healthy potato images. This not only expands the dataset but also enhances variety, aiding model generalization. Experimental results indicate that the images generated by PotatoGANs exhibit better quality and realism, closely resembling real disease images, as measured by the Inception score.
Specifically, the CycleGAN model outperforms the Pix2Pix GAN model in terms of image quality, achieving Inception scores of 1.2001 and 1.0900 for black scurf and common scab, respectively. The synthetic data significantly improves training for large neural networks, reduces data collection costs, and enhances data diversity and generalization capabilities.
Furthermore, our work enhances interpretability by integrating three gradient-based Explainable AI algorithms (GradCAM, GradCAM++, and ScoreCAM) with three distinct CNN architectures (DenseNet169, Resnet152 V2, InceptionResNet V2) for potato disease classification.
Blogger's Review: PotatoGANs offers a powerful tool for agricultural disease detection through innovative generative adversarial networks, significantly improving model robustness and accuracy, especially in data-scarce scenarios. This represents a notable advancement in the application of deep learning in agriculture.