Deep learning model for plant disease detection based on visual analysis of leaf infestation area

Abstract

The article considers modern methods based on deep learning for solving the problem of plant disease recognition. A comparative analysis of some existing methods is carried out. A modified neural network model is created that allows to surpass existing methods in recognition accuracy and memory costs. Using the VGG16 architecture, a modified convolutional model is developed using Keras. The dataset used were images of plants—tomato leaves, both healthy and diseased. A computational experiment was carried out in comparison with such architectures as VGG16, ResNet-50, and EfficientNet-85. The proposed model allows to detect plant diseases with the best results in computations and accuracy.

Plant phenotyping relevance

植物葉の病変領域を画像から認識する深層学習モデルの開発・比較が中心で、病害状態という植物表現型を直接推定しているため。

Code and data availability

The paper's Data availability statement points to the public PlantVillage plant image dataset (arXiv 1511.08060) used as the paper's phenotyping input. No author analysis code or trained model is publicly deposited.