Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

A novel approach to navigate the taxonomic hierarchy to address the Open-World Scenarios in Medicinal Plant Classification

Created by
  • Haebom

Author

Soumen Sinha, Tanisha Rana, Rahul Roy

Outline

In this paper, we present a novel approach for hierarchical classification of medicinal plants. Considering that existing methods have difficulty in hierarchical classification and identification of unknown species, we approach the problem of assigning optimal hierarchical labels to unknown species. We propose a novel method to perform hierarchical classification by integrating DenseNet121, multi-scale self-attention (MSSA), and cascade classifier. MSSA captures both local and global context information of the image to improve the performance of distinguishing between similar species and identifying new species. The proposed method shows excellent performance for both known and unknown species, and is evaluated on two state-of-the-art datasets with and without background artifacts. The accuracy for unknown species is 83.36% for phylum, 78.30% for class, 60.34% for order, and 43.32% for family. The model size is about 4 times smaller than existing state-of-the-art methods, making it easy to deploy in real-world applications.

Takeaways, Limitations

•
Takeaways:
◦
A novel approach to effectively solve the hierarchical classification problem of medicinal plants as an unknown species classification problem is presented.
◦
Achieving better performance than existing methods by integrating DenseNet121, MSSA, and cascading classifiers.
◦
Improving the performance of similar species identification and new species identification by utilizing multi-scale self-attention mechanisms.
◦
Ease of deployment in real-world applications is ensured by reducing the size of the model.
•
Limitations:
◦
Classification accuracy below the family level is relatively low (43.32%).
◦
Additional validation is needed on the scope of the dataset used and generalization performance.
◦
There may be a potential performance bias for certain medicinal plant species.
👍
Made with Slashpage