Multi-Class Plant Leaf Disease Detection Using a Deep Convolutional Neural Network

Abstract

Traditional machine learning methods of plant leaf disease detection lack successful performances due to poor feature representation and correlation. This paper presents a novel methodology for automatic plant leaf disease detection using cascaded deep convolutional neural network (CDCNN) which focusses on increasing the feature representation and correlation factors. It provides distinctive features that gives low intra-class variability and higher inter-class variability. CDCNN were performed on a plant-village leaf disease database which consists of 13 classes of tomato, potato, and pepper bell plant diseases; DCNN model performs better with an overall accuracy, recall, and precision of 98.50%, 0.98, and 0.97 respectively. Additionally, performance of the proposed algorithm is evaluated on real time cotton leaf database for bacterial blight, leaf miner, and spider mite diseases detection and provides 99.00% accuracy. The proposed DCNN outperforms well compared to traditional machine learning and deep learning models and is able to detect the diseases present in the leaves of the plant.

Plant phenotyping relevance

植物葉の病害状態を画像から推定する深層学習手法を開発・比較評価しており、植物フェノタイピング手法が中心である。

Code and data availability

The paper uses the PlantVillage dataset (cited prior work, not a paper-specific deposit) and a self-collected cotton leaf image set with no availability statement. No author code, models, or data URLs are provided.

No evidence-backed public reproduction asset is currently recorded.