Automated Plant Leaf Disease Diagnosis using Deep Learning

Abstract

Plant varieties are essential for the survival of human beings and animals, as they act as an alternative source of food, fiber, fodder, and other raw materials for domestic needs and industries in any society. Early identification of plant leaf diseases is very important for keeping the health of crops intact. Crops being infected can affect the overall yield of crops, which may be detrimental to the earnings of farmers. With the emergence of artificial intelligence technology, it has become possible to deploy systems for quicker identification of illnesses. This work has been carried out for the prediction of plant diseases based on visual phenotypic manifestations, such as images of leaves. For this purpose, the dataset has been created after retrieving data from the PlantVillage dataset. The clinical reliability of different deep learning models of various representational capacities has been tested while using ImageNet pre-trained parameters. The experimental results show that MobileNetV2 achieves the highest accuracy of 96.15%, outperforming deep CNN (76.92%) and medium CNN (61.54%). The test accuracy and class-wise F1-scores for the CNN are observed to be substantially very high. The generalization ability and result of DCNN and MCNN are moderate and poor, respectively, as observed. Additionally, the proposed CNN only uses the disease-affected areas on the leaf, thus making the result more interpretable

Plant phenotyping relevance

葉の画像から病徴を推定する深層学習手法を開発・比較し、モデル性能と解釈性を評価しているため、植物表現型取得が中心である。

Code and data availability

The paper trains CNN/MobileNetV2 models on a subset of the PlantVillage dataset, but no author code, trained models, or paper-specific data deposit is mentioned. The only public URL in the text (Kaggle Paddy Disease Classification) appears solely as cited reference [16] and is not the dataset used for this paper's own,

No evidence-backed public reproduction asset is currently recorded.