Upper and Lower Leaf Side Detection with Machine Learning Methods.

Abstract

Recent studies have approached the identification of foliar plant diseases using artificial intelligence, but in these works, classification is achieved using only one side of the leaf. Phytopathology specifies that there are diseases that show similar symptoms on the upper part of the leaf, but different ones on the lower side. An improvement in accuracy can be achieved if the symptoms of both sides of the leaf are considered when classifying plant diseases. In this context, it is necessary to establish whether the captured image represents the leaf on its upper or lower side. From the research conducted using botany books, we can conclude that a useful classification feature is color, because the sun-facing part is greener, while the opposite side is shaded. A second feature is the thickness of the primary and secondary veins. The veins of a leaf are more prominent on the lower side, compared to the upper side. A third feature corresponds to the concave shape of the leaf on its upper part and its convex shape on the lower part. In this study, we aim to achieve upper and lower leaf side classification using both deep learning methods and machine learning models.

Plant phenotyping relevance

葉の表裏という植物器官の状態を画像から分類する機械学習手法が研究の中心であり、病害画像解析に再利用可能な前処理・表現として植物フェノタイピング手法に該当する。

Code and data availability

The paper uses a self-collected sunflower leaf image dataset (field images taken with an iPhone, manually cropped/resized to 224×224) and ResNet/ML classifiers, but no block contains any public deposit, availability statement, URL, or supplement for the dataset, images, code, or trained models. All listed URLs are the

No evidence-backed public reproduction asset is currently recorded.