Disease classification in aubergine with local symptomatic region using deep learning models

Abstract

Recent trends in the application of deep learning techniques for crop disease classification is gaining international attention among experts of various domains. Many of the pioneering works have been carried out using the leaf images in laboratory condition with several shortcomings for implementation in field condition. Some of the other key issues are the presence of multiple disorders and the similarity of symptoms which can be addressed by using local symptomatic regions of the disease symptoms. In addition, although many studies discussed on the AI-based smartphone application for disease classification, very few studies have actually implemented it. This study has explored the classification of five diseases in Solanum melongena (also commonly known as eggplant, aubergine or brinjal) with the creation of the dataset consisting of local symptomatic region, utilised one of the popular deep learning model VGG16 for classification and optimised it for deployment in a smartphone. The VGG16 model was trained with fine-tuned hyperparameter and evaluated with a test dataset which resulted in the accuracy of 94.3%. This study also analysed the feature parameters from several layers using Multi-class Support Vector Machine (MSVM) to understand the learning process as it approaches top layers. Further, the feature parameters of dominant channels that significantly influenced the classification process were identified and analysed. Finally, VGG16 model was customised and implemented in a smartphone. It was tested in a trial condition which resulted in the classification accuracy of 91.3%. Discussions on the possible reasons for misclassification and scope for improvement have been provided.

Plant phenotyping relevance

ナス葉の局所的な病徴画像から病害状態を分類する画像解析手法を開発・最適化し、データセット、精度評価、スマートフォン実装まで扱っており、植物病害フェノタイピングが中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.