Related Experiment Video
Updated: Sep 7, 2025
05:16
Flying Insect Detection and Classification with Inexpensive Sensors
Published on: October 15, 2014
25.3K
A ResNet attention model for classifying mosquitoes from wing-beating sounds.
Xutong Wei1, Md Zakir Hossain2,3,4, Khandaker Asif Ahmed5
1Research School of Computing, Australian National University, Canberra, ACT, 2601, Australia.
Scientific Reports
|June 21, 2022
Summary
A new deep learning model, Wing-beating Network (WbNet), accurately classifies mosquito species using wing-beating sounds. This method offers a robust, field-deployable tool for mosquito control and disease prevention efforts.
Area of Science:
- Entomology
- Bioacoustics
- Machine Learning
Background:
- Mosquitoes transmit deadly diseases, making accurate classification crucial for control programs.
- Traditional classification methods are labor-intensive and time-consuming.
- Image-based machine learning models exist, but sound-based classification offers field advantages.
Purpose of the Study:
- To develop a deep neural network for classifying six mosquito species from three genera using wing-beating sounds.
- To enhance feature extraction by converting raw audio to Mel-spectrograms for improved robustness.
- To create a practical, field-deployable tool for mosquito identification.
Main Methods:
- Developed a deep neural network model named 'Wing-beating Network' (WbNet).
- Utilized a residual neural network (ResNet) baseline combined with self-attention and data augmentation.
- Pre-processed raw audio data into informative Mel-spectrograms for feature extraction.
Main Results:
- WbNet achieved high performance: 89.9% on WINGBEATS and 98.9% on ABUZZ datasets.
- Achieved 100% precision, recall, and F1-scores for Aedes and Culex genera.
- Demonstrated strong performance (>95%) for Anopheles, with no need for sophisticated audio equipment.
Conclusions:
- The WbNet model effectively classifies mosquito species based on wing-beating sounds.
- Mel-spectrograms provide robust, noise-free features for accurate classification.
- The model shows significant potential for mosquito monitoring, prevalence studies, and eradication programs.
