Related Experiment Video
Updated: Aug 3, 2026
10:56
Long-term Behavioral Tracking of Freely Swimming Weakly Electric Fish
Published on: March 6, 2014
12.5K
CLIB: Contrastive learning of ignoring background for underwater fish image classification
Qiankun Yan1,2, Xiujuan Du1,2,3, Chong Li1,2
1College of Computer, Qinghai Normal University, Xining, China.
Frontiers in Neurorobotics
|August 15, 2024
Summary
A new contrastive learning method ignoring background (CLIB) enhances underwater fish image classification by separating subjects from noise. This approach improves accuracy and robustness against complex backgrounds in aquatic environments.
Area of Science:
- Computer Vision
- Machine Learning
- Image Processing
Background:
- Underwater fish image classification faces challenges with background noise.
- Existing methods struggle with anti-interference capabilities in complex aquatic environments.
Purpose of the Study:
- To propose a novel contrastive learning method ignoring background (CLIB) for improved underwater fish image classification.
- Enhance the accuracy and robustness of fish image recognition by effectively handling background interference.
Main Methods:
- CLIB utilizes an extraction module to separate fish subjects from image backgrounds.
- A multi-view contrastive loss function is introduced to enhance subject-background distinction.
- Three complementary views are composed with the original image for contrastive learning.
Main Results:
- CLIB demonstrated significant performance improvements across multiple public datasets (Fish4Knowledge, Fish-gres, WildFish-30, QUTFish-89).
- Improvements ranged from 1.43% to 14.82% compared to baseline models.
- The method effectively focuses on core subject features, ignoring background noise.
Conclusions:
- CLIB offers a robust solution for underwater fish image classification.
- The proposed method significantly enhances classification accuracy and resilience to background noise.
- CLIB validates the effectiveness of subject-background separation in contrastive learning for aquatic imagery.
