A Novel Computer Vision Model for Medicinal Plant Identification Using Log-Gabor Filters and Deep Learning Algorithms

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Computer vision systems aim to mimic human visual perception.
• Deep learning, particularly Convolutional Neural Networks (CNNs), has advanced computer vision.
• Identifying medicinal plants is crucial for traditional medicine and drug discovery.

Purpose of the Study:

• To develop an accurate computer vision system for identifying medicinal plants using leaf textural features.
• To enhance Convolutional Neural Network (CNN) performance by integrating Log-Gabor filters.
• To evaluate the proposed system on diverse plant datasets.

Main Methods:

• Developed OTAMNet by fusing a Log-Gabor filter layer into the DenseNet201 architecture.
• Utilized transfer learning with ten pre-trained CNNs as feature extractors.
• Tested the system on a custom dataset (MyDataset) and four benchmark datasets (Flavia, Swedish Leaf, MD2020, Folio).

Main Results:

• The proposed OTAMNet model achieved 98% accuracy on the MyDataset.
• OTAMNet demonstrated high performance on benchmark datasets: Flavia (99%), Swedish Leaf (100%), MD2020 (99%), and Folio (97%).
• A false-positive rate below 0.1% was consistently achieved across all tested datasets.

Conclusions:

• The integration of Log-Gabor filters with DenseNet201 significantly improves medicinal plant identification accuracy.
• OTAMNet offers a robust and highly accurate solution for automated medicinal plant recognition.
• The system shows potential for applications in ethnobotany, pharmacology, and biodiversity monitoring.