DIMPSAR: Dataset for Indian medicinal plant species analysis and recognition

Abstract

Mobile-captured images of medicinal plants are widely used in various research investigations. Machine vision-based tasks such as the identification of plant species types for intelligent imaging device applications take a significant part in it. Botanists, farmers and researchers can reliably identify medicinal plants with the help of images captured using smartphones. Mobile captured images can be used for quality control to make sure that the right plant species are being used in pharmaceutical products. In the field of education, pictures of medicinal plants and their usage can be used to educate learners, medical professionals, and the general public. Further, various research investigations in the area of chemistry, pharmacology, the therapeutic potential of medicinal plants, images can be employed. In this paper, we contribute a dataset of Indian medicinal plant species. The dataset is collected from different regions of Karnataka and Kerala. Datasets include characteristics such as multiple resolutions, varying illuminations, varying backgrounds, and seasons in the year. The datasets consist of 5900 images of forty plant species and single leaf images of eighty plant species consisting of 6900 samples obtained from real-time conditions using smartphones. The datasets contributed would be useful to researchers to investigate on development of algorithmic models based on image processing, machine learning, and deep learning concepts to educate about medicinal plants. The dataset can be accessed by anybody, without charge, at DOI:10.17632/748f8jkphb.2, 10.17632/748f8jkphb.3.

Keywords: Leaf images; Mobile captured images; Plant classification; Plant/leaf analysis; Whole plant images.

Fig. 1

Image samples captured in the varying background (a) Leaf- Shadow (b) Leaf- Occluded (c) Leaf- Varying background (d) Leaf- Light Illumination (e) Leaf- Bright sunlight.

Fig. 2

Samples of whole plant images captured in varying background

Fig. 3

Augmented samples geometrical intensity transformations (a) Original image (b) Low contrast images (c) High contrast images (d)Rotated images (e) Flipped images.