Smartphone image dataset for turmeric plant leaf disease from Bangladesh spice fields.

Abstract

Agriculture is key to sustaining life and economic development, and crops like turmeric are essential for everyday application and economic viability. Turmeric crops are very prone to foliar disease, which has a great impact on yield and quality. Early detection of the diseases is of great significance to farming practitioners since manual observation is generally time-consuming and unreliable. To surpass this challenge, a comprehensive dataset has been developed to facilitate the generation of an automatic disease recognition system. The dataset comprises 865 images of original turmeric leaves and 3496 images of augmented turmeric leaves, both infected and healthy, with four classes of diseases: aphid attack, blotch, leaf spot, and healthy leaves. All the leaves were captured from different angles to offer variability and clarity, with particular emphasis on high-quality and diversified data. Through this dataset, a precise and efficient identification process can be realized, which will aid agriculture practitioners in recognizing diseases at an early stage and reducing crop losses. This paper seeks to improve agricultural productivity, crop quality, and the overall growth and sustainability of the agricultural economy using state-of-the-art deep learning models, such as EfficientNetB7 and ResNet152, for precise and interpretable disease classification. The proposed approach achieves high accuracy, with EfficientNetB7 attaining 98.67 % and ResNet152 reaching 97.87 %. Additionally, this research lays the groundwork for scalable and affordable disease detection technology, allowing agricultural practitioners to maximize crop yield and achieve long-term food security using smart tools.

Plant phenotyping relevance

ターメリック葉の病害状態を画像から分類するデータセットと深層学習手法が研究の中心であり、植物病害フェノタイピングに該当する。

Code and data availability

The paper is a Data in Brief article describing a turmeric leaf disease image dataset (865 original and 3496 augmented smartphone images) collected by the authors, with the dataset publicly deposited on Mendeley Data. This is a paper-specific, publicly available plant image/phenotyping asset with a direct URL matching,