Efficient Deep Learning Unleashed: The EDUS-Net Revolution in Medical Plant Species Identification Using Deep Learning

Accurate identification of medicinal plant species is crucial for health care, requiring accurate recognition of therapeutic plant species. Traditional approaches are time-consuming, urging the exploration of faster alternatives. Deep learning, with its ability to unravel complex patterns, offers a promising solution to enhance plant identification accuracy. However, challenges persist in identifying Indian medicinal plant species, particularly with the drawbacks of heavyweight pre-trained models such as extended training times and limitations in real-time applications without GPU support. Addressing these challenges, we introduce EDUS-Net, a smart and efficient model that overcomes the drawbacks of heavyweight pre-trained models. It simplifies the training process and enhances prediction performance. By harnessing the power of EfficientNetB2, dilated convolution, upsampling with skip connections, and thoughtful layer freezing, EDUS-Net speeds up computations, preserves crucial details, and improves accuracy—all without needing a high-powered GPU. Focused on swiftly identifying Indian medicinal plant species, EDUS-Net outshines other models, boasting an impressive 99.43% accuracy. Highlighting its efficiency, EDUS-Net significantly cuts down training times by 22.7% compared to a basic EfficientNetB2 model (931 s). More impressively, it predicts outcomes in just 45 s on a standard computer, supporting 357 images with a swift 0.13 s for each prediction—and all of this without relying on a powerful GPU. This makes EDUS-Net a game-changer, revolutionizing the identification of medicinal plant species with its user-friendly efficiency in healthcare applications.