Hybrid CNN-LSTM Model for Indian Medicinal Plant Classification

Since ancient times, Ayurvedic/medicinal plants have been vital for healthcare and are fundamental in the creation of drugs and other medical therapies. In developed countries, Ayurvedic plants account for more than 25% of medicines, compared to developing countries where approximately 80% of people depend on these plants for primary healthcare. Typically, experts manually identify these plants, but it is a slow, subjective process that relies on their availability. Moreover, an inaccurate detection could have dire consequences for one’s health or even be fatal. A more reliable and practical approach is required to accurately identify medicinal plants. This research study presents a hybrid architecture of the convolutional neural network (CNN) and long short-term memory (LSTM) network for Indian medicinal plant classification. Features were extracted using CNN attributes, and dependencies were calculated and classified using LSTM attributes. DenseNet169 architecture was used to extract the features. The proposed model was trained and tested on images of five medicinal plant species from a publicly available dataset. The hybrid DenseNet169-LSTM model achieved 99.88% testing accuracy, surpassing five CNN models, while maintaining reduced parameters and model complexity, proving the practicality of this approach.