Classification of Indian Medicinal Plant Species Using Attention Module With Transfer Learning

Accurate plant species classification is an important task in the automatic diagnosis of medicinal plants because it increases the diagnostic capacity of subsequent processing. The key predictable characteristics of a medicinal plants are their form, color, and texture of its leaves. This publication presents an novel approach to classify medicinal plant species using attention-based deep convolutional neural networks. The proposed model has a convolution block attention module (CBAM) that achieved multi-class classification with 92.10% accuracy, 92.0% precision, 92.0% recall, and 92.0% F1-Score respectively. During the preprocessing stage, images are separated from their backgrounds using color channel processing and morphological operations. The proposed classification method is compared with state of art methods. Also the classification is done thorough transfer learning (Dense-Net 121 with accuracy of 99.66%, Vision transformer with accuracy of 97.0%, and Swin transformer with accuracy of 92.44%). The plant species are identified using 5945 images representing 40 different plant species from the DIMPSAR (Dataset for Indian medicinal plant species analysis and recognition) database, resulting in forty classes. In addition, research was conducted to investigate the influence of segmentation. The proposed model can be utilized to precisely categorize plants in computer-aided systems and to investigate therapeutic difficulties.