Related Experiment Video
Updated: Aug 5, 2026
08:34
Automated Charting of the Visual Space of Housefly Compound Eyes
Published on: March 31, 2022
A structured literature review of computer vision methods for insect identification
Sudha Cheerkoot-Jalim1, Camille Simon-Chane2, Zarine Cadersaib1
1Faculty of Information, Communication and Digital Technologies, University of Mauritius, Réduit, Mauritius.
Journal of Insect Science (Online)
|July 31, 2026
Summary
Automated insect identification using computer vision faces challenges like limited data and field variability. Deep learning shows promise, but real-world generalization requires better models and diverse datasets for accurate biodiversity monitoring and pest control.
Area of Science:
- Computer Vision
- Entomology
- Biodiversity Monitoring
Background:
- Automated insect identification is vital for biodiversity, pest management, and vector surveillance.
- Current practical deployment is hindered by data scarcity, field variability, and taxonomic complexity.
Purpose of the Study:
- To conduct a PRISMA-guided literature review of computer-vision methods for insect classification and identification.
- To assess the quality and scope of existing research in automated insect identification.
Main Methods:
- Systematic literature search on Web of Science (930 records retrieved).
- Quality assessment of 230 full-text articles using weighted criteria.
- Data extraction on taxonomy, methods, datasets, and performance metrics from 111 high-quality studies.
Main Results:
- Deep learning, particularly Convolutional Neural Network (CNN) variants like ResNet and EfficientNet, dominates the field.
- CNN-based and one-stage detectors outperform traditional methods; transformers show potential for limited data.
- Laboratory performance often fails to generalize to field conditions due to illumination, occlusion, and pose variations.
Conclusions:
- Advancements require comprehensive reporting, lifecycle-aware and domain-adapted models validated in the field.
- Development of diverse, standardized benchmarks and interpretable architectures for embedded systems is crucial.
- Addressing data scarcity and geographical bias in public datasets is essential for reproducibility and equitable benchmarking.
