SIMPD Net: A Novel Dataset for South Indian Medicinal Plants in Natural Conditions

Automated plant recognition in real-time environments is a challenging problem in both botanical taxonomy and computer vision. Existing databases of medicinal plants are typically prepared for specific classes of plants and created in lab environments with white backgrounds. Handcrafted feature engineering is inadequate for handling large-scale datasets consisting of unconstrained images. The existing datasets lack mobile-based plant images acquired in natural scenes, which vary greatly due to different contributors, cameras, areas, periods of the year, individual plants, and other factors. To address this, a balanced dataset has been collected, resulting in the creation of a new database of medicinal plants. This database, named the South Indian Medicinal Plants (SIMPD) Dataset, contains 2,000 images collected from the nearby areas of Tirunelveli. The SIMP dataset includes 20 different classes of medicinal plants, such as herbs, shrubs, climbers, wild plants, and trees, all taken in unconstrained environments. To demonstrate the effectiveness of the SIMP dataset, it was evaluated using the proposed SIMPD net with 20 classes, achieving an accuracy of 94.04% compared to other deep networks.