Dataset of Virginia Flue-cured Tobacco Leaf images based on stalk leaf position for classification tasks: A case of Tanzania.

Abstract

Nicotiana tabacum is a kind of plant cultivated for its leaves used for manufacturing medicine and cigarettes. With the common name, the Tobacco plant is grown in many countries including China, Indonesia, Malawi and Tanzania just to mention a few. Literatures suggest a technical gap in the proper identification of grade labels for various parts of the plant. In addition, manual grading has resulted in various gaps and biases. To mitigate this, a data-driven grading solution is necessary. However, relevant datasets to train grade classifiers from various countries become of the essence. This article presents images concentrated on tobacco leaf plant position namely Leaf position which normally carries 23 grade labels. Due to high rainfall which swiped away the applied fertilizer on the tobacco plants in the farms, we failed to get images of one grade. Therefore, this research could capture and label 22 grade labels. Images of tobacco leaves based on the tobacco plant position were collected in Tanzania through participatory community research. Canon 5D mark III cameras with 100 mm micro lens were used to take pictures of tobacco leaves based on the tobacco plant position. Domain experts were used for image labelling and cleaning according to tobacco grade labels identified in Tanzania. The dataset carries 49,779 images, which can be used to develop machine learning models for tobacco leaf grade label identification. The collected dataset can be used to train models and enhance the performance of pre-trained models in any country of interest.

Plant phenotyping relevance

タバコ葉の位置・等級を画像として収集し、専門家ラベル付きデータセットを構築しており、植物器官の状態・品質を画像から分類する再利用可能な方法資源が中心である。

Code and data availability

The paper is a data descriptor whose own tobacco leaf image dataset (49,779 images, 22 grade labels) is publicly deposited in the Harvard Dataverse with an explicit direct URL, making it a paper-specific, publicly actionable phenotyping image dataset.