Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning Domain-Robust Bioacoustic Representations for Mosquito Species Classification with Contrastive Learning and Distribution Alignment

Yuanbo Hou, Zhaoyi Liu, Xin Shen, Stephen Roberts

TL;DR

This work tackles domain shift in mosquito species classification from bioacoustic recordings, where models can overfit to non-biological domain cues. The authors introduce DR-BioL, a framework that fuses contrastive learning with species-cohesion and domain-invariant cues, plus species-conditional distribution alignment, to produce domain-robust representations. Through a CNN encoder and five targeted losses, including $\mathcal{L}_\mathrm{ScL}$, $\mathcal{L}_\mathrm{ScoL}$, $\mathcal{L}_\mathrm{SdaL}$, $\mathcal{L}_\mathrm{DcL}$, and $\mathcal{L}_\mathrm{DicL}$, the approach achieves improved cross-domain MSC performance on a multi-domain dataset spanning eight species and four environments. The results indicate DR-BioL maintains species cues while mitigating domain information, enhancing practical applicability for real-world vector surveillance.

Abstract

Mosquito Species Classification (MSC) is crucial for vector surveillance and disease control. The collection of mosquito bioacoustic data is often limited by mosquito activity seasons and fieldwork. Mosquito recordings across regions, habitats, and laboratories often show non-biological variations from the recording environment, which we refer to as domain features. This study finds that models directly trained on audio recordings with domain features tend to rely on domain information rather than the species' acoustic cues for identification, resulting in illusory good performance while actually performing poor cross-domain generalization. To this end, we propose a Domain-Robust Bioacoustic Learning (DR-BioL) framework that combines contrastive learning with distribution alignment. Contrastive learning aims to promote cohesion within the same species and mitigate inter-domain discrepancies, and species-conditional distribution alignment further enhances cross-domain species representation. Experiments on a multi-domain mosquito bioacoustic dataset from diverse environments show that the DR-BioL improves the accuracy and robustness of baselines, highlighting its potential for reliable cross-domain MSC in the real world.

Learning Domain-Robust Bioacoustic Representations for Mosquito Species Classification with Contrastive Learning and Distribution Alignment

TL;DR

This work tackles domain shift in mosquito species classification from bioacoustic recordings, where models can overfit to non-biological domain cues. The authors introduce DR-BioL, a framework that fuses contrastive learning with species-cohesion and domain-invariant cues, plus species-conditional distribution alignment, to produce domain-robust representations. Through a CNN encoder and five targeted losses, including , , , , and , the approach achieves improved cross-domain MSC performance on a multi-domain dataset spanning eight species and four environments. The results indicate DR-BioL maintains species cues while mitigating domain information, enhancing practical applicability for real-world vector surveillance.

Abstract

Mosquito Species Classification (MSC) is crucial for vector surveillance and disease control. The collection of mosquito bioacoustic data is often limited by mosquito activity seasons and fieldwork. Mosquito recordings across regions, habitats, and laboratories often show non-biological variations from the recording environment, which we refer to as domain features. This study finds that models directly trained on audio recordings with domain features tend to rely on domain information rather than the species' acoustic cues for identification, resulting in illusory good performance while actually performing poor cross-domain generalization. To this end, we propose a Domain-Robust Bioacoustic Learning (DR-BioL) framework that combines contrastive learning with distribution alignment. Contrastive learning aims to promote cohesion within the same species and mitigate inter-domain discrepancies, and species-conditional distribution alignment further enhances cross-domain species representation. Experiments on a multi-domain mosquito bioacoustic dataset from diverse environments show that the DR-BioL improves the accuracy and robustness of baselines, highlighting its potential for reliable cross-domain MSC in the real world.

Paper Structure

This paper contains 13 sections, 9 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Domain-Robust Bioacoustic Learning
  3. Bioacoustic representation encoder
  4. Species classification loss (ScL)
  5. Domain classification Loss (DcL)
  6. Contrastive cross-domain species cohesion loss
  7. Species-conditional distribution alignment loss
  8. Total loss
  9. Experiments and results
  10. Dataset, experiments setup, and metrics
  11. Results and analysis
  12. CONCLUSION
  13. ACKNOWLEDGEMENTS

Figures (4)

  • Figure 1: Spectrograms from different sources show that the CNN with illusory high test accuracy in Table \ref{['tab:cnn_demo']} classifies Aedes albopictus samples by domain features of D2 rather than species information of Aedes albopictus.
  • Figure 2: A CNN-based example of an instantiation of the proposed DR-BioL framework.
  • Figure 3: Distribution of wingbeat frequencies for the mosquito species data used in this paper.
  • Figure 4: Visualization of the domain embeddings using t-SNE.