Related Experiment Video
Updated: May 11, 2026
05:16
Flying Insect Detection and Classification with Inexpensive Sensors
Published on: October 15, 2014
25.1K
UB-Former: A fine-grained classification method for images of insects using biomorphic features
Shilu Kang1, Hua Huo1, Aokun Mei1
1School of Information Engineering, Henan University of Science and Technology, Luoyang, 471023, Henan, China.
Computational Biology and Chemistry
|May 15, 2025
Summary
This study introduces UB-Former, a novel model for fine-grained insect image classification using biomorphic features. It achieves state-of-the-art results by effectively isolating targets and comparing features across life stages.
Area of Science:
- Computer Vision
- Machine Learning
- Entomology
Background:
- Fine-grained insect image classification demands precise identification of species-specific traits across diverse life stages.
- Existing methods often struggle with background noise and variations in insect appearance throughout their development.
Purpose of the Study:
- To develop a novel fine-grained insect image classification model, termed UB-Former, leveraging biomorphic information.
- To enhance classification accuracy by addressing challenges in target isolation and feature extraction.
Main Methods:
- A segmentation module isolates insect targets, reducing background interference.
- The using biomorphic features module (UBM) extracts dual-channel features from segmented images and contour textures.
- A multi-image feature comparison method (MIFC) facilitates cross-domain learning for improved performance.
Main Results:
- UB-Former achieved state-of-the-art accuracy of 61.1% on the Insecta dataset and 53.2% on the IP102 dataset.
- The model demonstrated high accuracy (92.0%-95.0%) on other fine-grained classification datasets like CUB200-2011, Stanford Cars, and Stanford Dogs.
- These results highlight the model's robustness and effectiveness across different domains.
Conclusions:
- The proposed UB-Former model effectively addresses challenges in fine-grained insect image classification.
- Its ability to capture shared features across life stages and domains signifies a significant advancement in the field.
