Analysis of Leaf's disease detection by improving the image quality using super resolution generative adversial networks

Abstract

In most cases, the acquired pictures for agricultural disease image identification are not very clear, which might result in subpar identification outcomes when used in actual production settings. The identification accuracy of pre-trained image classifiers is greatly affected by the picture quality. We suggest DATFGAN, a generative adversarial network that combines topology-fusion with dual-attention, to solve this issue. With this network, even low-resolution photographs may be improved to a much higher standard. Our suggested network also has a weight sharing method that may drastically cut down on parameters. The experimental findings show that compared to state-of-the-art approaches, DATFGAN produces better outcomes in terms of visual appeal. Furthermore, identification tasks are used to assess the altered pictures. The findings show that the suggested strategy is strong enough for real-world applications and performs far better than competing methods.