Issue |
ITM Web Conf.
Volume 70, 2025
2024 2nd International Conference on Data Science, Advanced Algorithm and Intelligent Computing (DAI 2024)
|
|
---|---|---|
Article Number | 03017 | |
Number of page(s) | 11 | |
Section | Image Processing and Computer Vision | |
DOI | https://doi.org/10.1051/itmconf/20257003017 | |
Published online | 23 January 2025 |
Exploring CycleGAN Technique for Improved Plant Disease Detection and Analysis
Olin Business School, Washington University in St. Louis, St. Louis, MO, 63l30, United States of America
The threat of plant disease is a significant problem affecting the world, when untreated these diseases can affect food production. Diagnosis of these diseases in an un-delayed manner is very important, however, methods described in current use that only involve the use of sight are inefficient and are also subject to errors. This paper tackles the problem by using Cycle-Consistent General Adversarial Networks (CycleGAN) to create artificial images of diseased plant leaves. The advantage of this approach is that augmenting the training data with images that do not exist in the real world helps improve the performance of disease classifications. The research takes into consideration the apple leaves diseased images, is of various pathogens, and CycleGAN creates images to even it. The results indicate that CycleGAN is indeed able to generate artificial images for the less complicated sicknesses associated with a mere shift in color, with an achieved micro-average Area Under the Curve (AUC) of .98 and macro-average AUC of 0.94. On the contrary, this model has problems in striking a balance while dealing with more complex diseases that have problems that are underlying structural deformation. However, adding such images in training datasets increases the classification accuracy in total. Future work should involve making the model more robust to complex and rich visual details as well as employing more sophisticated models for better applicability in real farming settings.
© The Authors, published by EDP Sciences, 2025
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.