A Learning Robust and Discriminative Shape Descriptor for Plant Species Identification

Area of Science:

• Botany and Computer Vision
• Artificial Intelligence applications in biological sciences

Background:

• Accurate plant identification from leaf images is crucial in botany and AI.
• Extracting robust, discriminative features for similarity measurement remains a challenge.

Purpose of the Study:

• To develop an effective and robust shape descriptor for plant species identification using leaf images.
• Introduce the High-Level Triangle Shape Descriptor (HTSD) for improved plant recognition.

Main Methods:

• Extracting external leaf contours and internal salient point information.
• Developing Contour Point based Triangle Features (CPTFs) and Salient Point Triangle Features (SPTFs).
• Encoding local triangle features using Fisher vectors to create the HTSD and using Euclidean distance for similarity.

Main Results:

• Extensive evaluation on public leaf datasets demonstrated successful application.
• The proposed HTSD method achieved superior recognition accuracy.
• Outperformed current state-of-the-art shape-based and deep-learning plant identification techniques.

Conclusions:

• The High-Level Triangle Shape Descriptor (HTSD) offers a robust and accurate solution for plant identification.
• This shape-based approach provides a competitive alternative to deep learning methods in plant recognition.