Crop Leaf Disease Image Super-Resolution and Identification With Dual Attention and Topology Fusion Generative Adversarial Network

Qiang Dai, Xi Cheng, Yan Qiao, Youhua Zhang

Abstract: For agricultural disease image identification, obtained images are typically unclear, which can lead to poor identification results in real production environments. The quality of an image has a significant impact on the identification accuracy of pre-trained image classifiers. To address this problem, we propose a generative adversarial network with dual-attention and topology-fusion mechanisms called DATFGAN. This network can effectively transform unclear images into clear and high-resolution images. Additionally, the weight sharing scheme in our proposed network can significantly reduce the number of parameters. Experimental results demonstrate that DATFGAN yields more visually pleasing results than state-of-the-art methods. Additionally, treated images are evaluated based on identification tasks. The results demonstrate that the proposed method significantly outperforms other methods and is sufficiently robust for practical use.