Iterative Super Resolution Network (ISNR) for Potato Leaf Disease Detection

Abstract

Since ancient times, agricultural diseases brought on by pests, bacteria, viruses, and fungus have caused significant food loss that requires worldwide attention. Therefore, crop disease diagnosis as early as possible can significantly prevent loss of yield as well increase monetary value. The disease in crop may be identified by carefully analysing either a leaf, node, or stem. Here, accurate disease diagnosis will typically depend on the resolution of the image. Iterative Super-Resolution Network (ISNR) model is used for analysing low resolution potato leaf and identifying the disease. Through a stochastic iterative denoising procedure, ISNR accomplishes super-resolution while adjusting denoising diffusion probability models by image to image translation. The presented ISNR model is evaluated using the publicly accessible PlantVillage dataset with super resolution factors 2, 4, and 6. For super resolution factors 2, 4, and 6, our model gets PSNR 33.781 dB, 35.292 dB, 37.538 dB, SSIM 0.817, 0.892, 0.953, and classification accuracies of 99.61, 98.05, and 96.09 respectively.