BDHerbalPlants: augmented and curated herbal plants image dataset for classification

Area of Science:

• Botany
• Computer Science
• Agricultural Informatics

Background:

• Accurate identification of herbal plants is crucial for agricultural research, healthcare, and pharmaceutical development.
• Existing datasets may lack diversity or expert-verified labeling, hindering the development of robust identification tools.

Purpose of the Study:

• To introduce the BDHerbalPlants dataset, a comprehensive collection of high-quality images of eight distinct herbal plant species.
• To demonstrate the utility of the BDHerbalPlants dataset in training and evaluating deep learning models for plant identification.

Main Methods:

• Collected 1792 raw, high-quality images and 14,336 augmented images of eight specific herbal plants (Eclipta prostrata, Ocimum tenuiflorum, Centella asiatica, Mentha arvensis, Kalanchoe pinnata, Azadirachta indica, Coriandrum sativum, Datura stramonium).
• Ensured expert verification and labeling for all images.
• Utilized pre-trained deep learning models (Xception, DenseNet201, RegNetY032) to showcase the dataset's effectiveness.

Main Results:

• The BDHerbalPlants dataset comprises 1792 raw and 14,336 augmented images across eight herbal plant species.
• Demonstrated successful application of the dataset with popular deep learning models, indicating its suitability for training.
• The dataset facilitates accurate classification and identification of herbal plants.

Conclusions:

• The BDHerbalPlants dataset is a valuable resource for advancing agricultural informatics and plant identification.
• Its integration into deep learning tasks can significantly benefit the healthcare and pharmaceutical industries.
• Facilitates research in classifying challenging-to-identify wild herbal plants.