Hybrid CNN-PSO Approach for Accurate Classification of Medicinal Flowers

Abstract

Accurate classification and identification of flowering medicinal plants hold immense potential for advancing traditional medicine and biodiversity conservation. Currently, experts manually identify medicinal plants based on characteristics such as aroma, flowers, and visual features—a process that is not only time-consuming but also prone to human error and inefficiencies. To address these challenges, this study proposes a robust medicinal flower classification framework composed of three major phases: segmentation, feature extraction, and classification. The segmentation phase utilizes a deep learning-based Detectron2 model to isolate flower regions accurately. Three distinct feature extraction schemes are then applied to capture the color and texture patterns of the segmented flowers. The extracted features are optimized and classified using a Convolutional Neural Network (CNN) enhanced with an Improved Seeding Particle Swarm Optimization (IS-PSO) algorithm. Dominant features are selected using a Particle Swarm Optimization (PSO) technique prior to classification to improve computational efficiency and model accuracy. The proposed method was validated using 14 real-world medicinal flower datasets, all of which hold significant medicinal value. The CNN-IS-PSO model achieved superior performance, with an accuracy of 96.24%, precision of 96.90%, recall of 93.41%, and an F1-score of 94.69%, significantly outperforming recent state-of-the-art models. This method demonstrates the potential to enhance the reliability and accuracy of automated medicinal plant classification.