Bio-optimized complex valued spatiotemporal GNN for herbal species classification.

Area of Science:

• Computer Vision
• Machine Learning
• Botany

Background:

• Herbal medicine relies on accurate plant identification, which is currently labor-intensive.
• Thousands of herb species exist, making manual classification challenging and time-consuming.
• Automated systems are needed to streamline the process of herbal species identification.

Purpose of the Study:

• To develop an automated computer vision system for accurate herb classification.
• To create a robust and optimized classification model using graph neural networks.
• To leverage morphological features of herbal leaves for improved identification.

Main Methods:

• Proposed a novel automated herb classification via complex-valued spatiotemporal graph convolutional neural network (AHC-CVSTGCN).
• Utilized Multiple Local Particle Filter (MLPF) for image preprocessing and Revised Tunable Q-Factor Wavelet Transform (RQFWT) for feature extraction (shape, color, texture).
• Employed hierarchical manta ray foraging optimization (HMRFO) to enhance the classification accuracy of the CVSTGCN model.

Main Results:

• The AHC-CVSTGCN model achieved a high accuracy of 99.40%.
• The system demonstrated excellent performance with 99.11% precision and 99.12% recall.
• Experimental results significantly outperformed existing methods in automated herbal species classification.

Conclusions:

• The developed AHC-CVSTGCN system offers a reliable and effective solution for automated herbal species classification.
• This automated approach can significantly reduce the time and labor involved in identifying medicinal plants.
• The study presents a paradigm shift, empowering medicinal plant science and healthcare practitioners with informed decision-making tools.