Automatic plankton image classification combining multiple view features via multiple kernel learning.

Area of Science:

• Marine Biology
• Ecosystem Monitoring
• Image Classification

Background:

• Plankton are crucial to marine food webs and sensitive to environmental changes.
• Existing plankton classification systems lack broad applicability and accuracy across devices.
• A practical, automated system for plankton image classification is needed.

Purpose of the Study:

• To develop an accurate and widely applicable automatic plankton image classification system.
• To address limitations of existing systems tied to specific imaging devices and narrow taxonomic scopes.

Main Methods:

• Proposed an automatic plankton image classification system combining multiple view features via multiple kernel learning (MKL).
• Integrated general and robust features, including Inner-Distance Shape Context for morphology.
• Utilized feature selection and MKL to optimally combine features from different views and datasets.

Main Results:

• The system was implemented on three datasets covering over 20 plankton categories.
• Experimental results demonstrated superior accuracy and robustness compared to state-of-the-art systems.
• The approach effectively handles diverse plankton images from various imaging devices.

Conclusions:

• An automatic plankton image classification system using multiple view features and MKL was successfully demonstrated.
• Combining multiple features with MKL significantly enhances classification accuracy.
• The system provides a valuable tool for understanding marine ecosystems and environmental changes.