A parallel convolutional neural network with background removal and lesion segmentation for field plant disease severity classification

Abstract

Today, the intelligent automation of agriculture has received much attention from researchers. One of the important factors for the success of this automation is the timely diagnosis of plant disease and making a decision appropriate to the existing conditions of the plant. Since the progress of the disease is a determining factor in the type of treatment method, the diagnosis of the severity of the disease is of particular importance. However, accurate diagnosis of plant disease progression depends on various factors, including the availability of appropriate and well-annotated training datasets for designing an efficient diagnostic system. On the other hand, the similarity of the complications of different diseases has made this work challenging. In this study, two tomato diseases, namely Bacterial Spot and Mosaic Virus, are investigated using images collected from the PlantVillage, Taiwan tomato leaves, Field-PlantVillage, and Syn-PlantVillage datasets. The disease severity levels are divided into six stages for Bacterial Spot and four stages for Mosaic Virus, and a specifically designed deep convolutional neural network is proposed for severity classification. Experimental results demonstrate that the proposed method achieves high accuracy under challenging field conditions and outperforms several state-of-the-art methods.

Plant phenotyping relevance

トマト葉の病徴・病害重症度を画像から段階分類するCNN、背景除去、病斑セグメンテーションを開発しており、植物状態の取得・推定手法が中心である。

Code and data availability

The paper's own severity-annotated datasets are explicitly restricted (available only on request), so no public paper-specific data asset qualifies. The authors do provide an explicit public code availability link for their proposed BaSPaC model. The Mendeley and Drive links are pre-existing external datasets cited as,