Earticle

초록

영어

Fire blight is a kind of bacterial disease, which particularly gives serious damage to apples and pears. There is no clear cure for fire blight until now and its infectious speed is fast. Thus, damage due to fire blight should be minimized through early diagnosis. With the development of artificial intelligence in recent years, deep learning has been widely used in the agricultural field. As already known, a deep learning model needs a large number of training datasets. However, fire blight does not occur frequently. Thus, the number of their datasets is very insufficient. To increase this insufficient number of datasets, a data augmentation method in relation to fire blight has been previously conducted but it failed to accurately generate images of fire blight symptoms. In this study, CycleGAN was used to generate accurate fire blight leaf images, and an unpaired dataset, which was used previously by default, was converted into a paired dataset, in which leaves were placed in the same direction. As a result, accurate fire blight leaf images were still not generated when an unpaired dataset was used, but when a paired dataset was used, images with accurate fire blight symptoms were generated.

목차

Abstract
I. INTRODUCTION
II. RELATED WORK
A. Studies on disease and pest image augmentation using GAN algorithm
B. Studies on disease and pest image augmentation using a hybrid technique
III. DATASET AND PRE-PROCESSING
IV. EXPERIMENT DESIGN AND RESULTS
V. CONCLUSION
REFERENCES

저자

• Ri Zheng [ Department of Computer and Engineering Department of Convergence Engineering for Intelligent Drone Sejong University ]
• HeLin Yin [ Department of Computer and Engineering Sejong University ]
• Dong Jin [ Department of Computer and Engineering Department of Convergence Engineering for Intelligent Drone Sejong University ]
• JiMin Lee [ Department of Computer and Engineering Department of Convergence Engineering for Intelligent Drone Sejong University ]
• Yeong Hyeon Gu [ Department of Computer and Engineering Sejong University ] Corresponding Author
• Seong Joon Yoo [ Department of Computer and Engineering Sejong University ]

참고문헌