文献类型： 外文期刊

作者： Chen, Jiasheng ¹ ; Li, Hongwei ¹ ; Zhang, Shunsheng ² ; Wu, Tao ³ ;

作者机构： 1.Guangxi Univ, Sch Mech Engn, Nanning 530004, Peoples R China

2.Yunnan Acad Agr Sci, Sugarcane Res Inst, Kaiyuan 661699, Peoples R China

3.South China Agr Univ, Coll Engn, Guangzhou 510642, Peoples R China

4.Guangxi Univ, Sch Mech Engn, Guangxi Key Lab Mfg Syst & Adv Mfg Technol, Nanning 530004, Peoples R China

关键词： Sugarcane leaf disease identification; Deep learning; YOLOv10n-seg-p6; CycleGAN; Mobile application

期刊名称：SUGAR TECH （ 影响因子：2.0； 五年影响因子：2.0 ）

ISSN： 0972-1525

年卷期： 2025 年

页码：

收录情况： SCI

摘要： Sugarcane, an important tropical/subtropical crop, suffers yield and quality losses from foliar diseases. Traditional manual identification methods are inefficient and often lead to overuse of pesticides. Recent advancements in deep learning, especially convolutional neural networks, show promise for plant disease recognition. However, challenges such as limited field data and complex backgrounds persist. This research proposes an integrated "YOLOv10n-seg-p6 + CycleGAN" framework for mobile-based sugarcane disease identification. First, YOLOv10n-seg-p6 is employed to accurately segment sugarcane leaf regions and eliminate background interference; then, the segmented pure leaf images are fed into CycleGAN to transform healthy leaves into diseased ones with realistic pathological features, thereby achieving high-quality data augmentation and sample balancing to enhance the performance of the classification model. Experimental results demonstrate enhanced authenticity of synthetic samples compared to traditional CycleGAN approaches, with the average FID score decreasing by 4.61 (from 74.85 to 70.24). Swin Transformer classifier trained on enhanced data achieves 98.2% test accuracy, outperforming models trained on original or standard CycleGAN-generated data. The mobile solution, tested under 30 concurrent user requests, ensures stable field deployment, offering efficient, real-time disease diagnosis for farmers.