An Effective Medicinal Plant Classification with Cascaded Network

Medicinal plants have played a pivotal role in healthcare throughout history, contributing to the development of numerous drugs and medical treatments. In affluent countries, over 25% of medicines are derived from medicinal plants, while approximately 80% of individuals in developing nations rely on these plants for primary healthcare. Traditionally, the identification of medicinal plants required manual assessment by specialists, a process that was laborious, time-consuming, subjective, and reliant on the availability of experts. Moreover, incorrect diagnoses could lead to severe health complications or even fatalities. This research introduces a practical, accurate, and reliable method for identifying medicinal plants using images captured with a smartphone. The proposed approach involved feature extraction using a pre-trained ResNet50 model within a cascaded architecture. These features underwent further refinement through Particle Swarm Optimization (PSO) to aid in the classification of plants using a Support Vector Machine (SVM). The resulting ResNet50-PSO-SVM network achieved an impressive accuracy rate of 99.60% in categorizing seven medicinal plants, surpassing the state-of-the-art accuracy of 99%. Moreover, the system demonstrated remarkable efficiency, capable of classifying an image in just 0.15 seconds, with an average accuracy of 97.79%. Extensive testing on three different cellphones confirmed the practical application of the system, highlighting its resilience, rapid identification process, and exceptional accuracy. These findings underscore the necessity for more accurate and effective techniques in identifying medicinal plants. While previous research has proposed numerous automated methods leveraging both deep learning and traditional machine learning (TML) approaches