Slice Residual U-Net Based Rice Plant Disease Classification Using Convolutional Attentional BiGRU

Abstract

Rice is one of the highest-produced staples all over the world. The production of rice is frequently threatened by leaf diseases in rice plants, which impacts farmers. This has an impact on economic growth in general. As a result, prompt diagnosis is essential, and appropriate action should be taken to enhance rice yield. The proposed framework in this article introduces a novel deep learning-based method named Convolutional Attentional Bidirectional Gated Recurrent Unit (CAtt_BiGRU) to classify rice leaf diseases. First, pre-processing techniques such as image rescaling and upgraded wiener filtering are applied to the input rice plant leaf images. After pre-processing, slice-based residual U-Net is used to segment the affected regions. The segmented images are then fed to the classification framework. The CAtt_BiGRU model was employed to classify diseases in rice plants. A Convolutional Neural Network (CNN) was employed for feature extraction, and the classification task was carried out by a Bidirectional Gated Recurrent Unit (BiGRU). Our approach surpasses other state-of-the-art methods by achieving an accuracy of 99.64%.

Plant phenotyping relevance

イネ葉の病害領域を画像からセグメンテーションし、病害を分類する手法が研究の中心であり、植物の病害状態を直接推定するため、植物フェノタイピング手法として収載する。

Code and data availability

The paper uses a public Kaggle rice leaf disease image dataset as its phenotyping input; no author code or model release is stated.