Efficacy of chili plant diseases classification using deep learning: a preliminary study

Abstract

Plant disease classification using deep learning techniques is a popular research area due to the numerous opportunities for introducing advance and robust classifiers. Nevertheless, classifying chilli plant diseases accurately from images under uncontrolled environment and various imaging conditions remains unsolved due to the lack of chilli disease image datasets. In this study, the efficacy of three high-performance deep learning algorithms, namely VGG16, InceptionV3, and EfficientNetB0, in classifying three types of chilli leaves diseases, namely upward curling, mosaic/mottling, and the bacterial spot, is demonstrated. These methods are popularly used for other plant disease classifications due to their effectiveness. The experiments were performed on the 3,000 chilli plant disease images collected from three different field environments in Selangor, Malaysia. The images were captured with a complex background and various illuminations, angles, and distances to reflect the real-life scenarios. The complexity of the collected images was created based on the taxonomic information of chilli leaves diseases and the unavailability of chilli disease images under various imaging conditions in the publicly available plant disease databases. Experimented using appropriate specifications, the models demonstrated outstanding performance with more than 95% accuracy with the highest accuracy of 98.83% by InceptionV3.

Plant phenotyping relevance

唐辛子葉の画像から病害状態を深層学習で分類する手法を開発・比較し、異なる野外画像条件で性能評価しているため、植物フェノタイピング手法が中心です。

Code and data availability

The paper uses a self-collected dataset of 3,000 chili disease images and standard Keras pre-trained models (VGG16, InceptionV3, EfficientNetB0), but no public dataset deposit, author code, trained checkpoints, or supplementary assets are mentioned anywhere in the supplied blocks. No availability statement or URL for a

No evidence-backed public reproduction asset is currently recorded.