BDHerbalPlants

Herbal plants are naturally abundant and offers a wide range of healthy medicinal effect. These are used as remedy for many common and fatal diseases. Manually identifying these are labor intensive and requires specific knowledge. Thus, automatic identification of these plants in allow everyone to use them hence is the goal of this dataset.

This dataset holds eight of these common yet very useful medicinal herbs for people to easily classify. The eight plants included are Kalokeshi (Eclipta prostrata), Tulsi (Ocimum tenuiflorum), Thankuni (Centella asiatica), Pudina (Mentha arvensis), Pathor Kuchi (Kalanchoe pinnata), Neem (Azadirachta indica), Dhonya Pata (Coriandrum sativum) and Datura (Datura stramonium). The data is mainly for plant identification purpose, it includes field-view images of all the plants. The data are primarily collected from National Botanical Garden and other data from Kumudini Women’s Medical College.

The “BDHerbalPlants” dataset holds 1792 raw images (as well as resized raw image) and 14336 augmented, pre-processed images of herbal plants. The raw images are further got split into 3 folders: Train, Test and Validation with corresponding ratio of 70%, 20% and 10% while resized raw images are not kept intact. Among the raw data 1251 are processed train images and keeping 187 images for validation while testing the results on 354 test data. Augmentation technique include Rotation, Scaling, Elastic Transformation, Gaussian Blur, Brightness Adjustment, Contrast Adjustment, Coarse Dropout, Horizontal Flip, Vertical Flip.

BDHerbalPlants Dataset will help the interested researchers further use this for machine learning and deep learning task for agriculture and botany. A diverse and high-quality dataset is definitely going to be helpful for sustainable agricultural practices promoting efficiency and reducing cost.

1. Original Dataset (Raw):
Number of datasets: 1792
Data format: .jpg

2. Augmented Dataset:
Number of datasets: 14336
Data format: .jpg

3. Original Dataset Resized (Raw Resized):
Number of datasets: 1792
Data format: .jpg