2019/11/25 by Takumi Saikawa, Quan Huu Cap, Saikawa, Takumi +7 · 2 citations
Agricultural and Biological Sciences · Chemistry · #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Plant Virus Research Studies #Smart Agriculture and AI #Spectroscopy and Chemometric Analyses
paper · pdf · doi:10.48550/arxiv.1911.10727
openalex publication_date 2019/11/25 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
In image-based plant diagnosis, clues related to diagnosis are often unclear,\nand the other factors such as image backgrounds often have a significant impact\non the final decision. As a result, overfitting due to latent similarities in\nthe dataset often occurs, and the diagnostic performance on real unseen data\n(e,g. images from other farms) is usually dropped significantly. However, this\nproblem has not been sufficiently explored, since many systems have shown\nexcellent diagnostic performance due to the bias caused by the similarities in\nthe dataset. In this study, we investigate this problem with experiments using\nmore than 50,000 images of cucumber leaves, and propose an anti-overfitting\npretreatment (AOP) for realizing practical image-based plant diagnosis systems.\nThe AOP detects the area of interest (leaf, fruit etc.) and performs brightness\ncalibration as a preprocessing step. The experimental results demonstrate that\nour AOP can improve the accuracy of diagnosis for unknown test images from\ndifferent farms by 12.2% in a practical setting.\n