BioDCASE 2026 Challenge Baseline for Cross-Domain Mosquito Species Classification

Yuanbo Hou; Vanja Zdravkovic; Marianne Sinka; Yunpeng Li; Wenwu Wang; Mark D. Plumbley; Kathy Willis; Stephen Roberts

eess.ASMar 20, 2026

BioDCASE 2026 Challenge Baseline for Cross-Domain Mosquito Species Classification

Authors: Yuanbo Hou, Vanja Zdravkovic, Marianne Sinka, Yunpeng Li, Wenwu Wang, Mark D. Plumbley, Kathy Willis, Stephen Roberts

Organizations: Machine Learning Research Group, University of Oxford · Oxford Long-Term Ecology Laboratory, University of Oxford · King’s College London · Centre for Vision, Speech and Signal Processing, University of Surrey

Abstract

Mosquito-borne diseases affect more than one billion people each year and cause close to one million deaths. Traditional surveillance methods rely on traps and manual identification that are slow, labor-intensive, and difficult to scale. Audio-based mosquito monitoring offers a non-destructive, lower-cost, and more scalable complement to trap-based surveillance, but reliable species classification remains difficult under real-world recording conditions. Mosquito flight tones are narrow-band, often low in signal-to-noise ratio, and easily masked by background noise, and recordings for several epidemiologically relevant species remain limited, creating pronounced class imbalance. Variation across devices, environments, and collection protocols further increases the difficulty of robust classification. Such variation can cause models to rely on domain-specific recording artefacts rather than species-relevant acoustic cues, which makes transfer to new acquisition settings difficult. The BioDCASE 2026 Cross-Domain Mosquito Species Classification (CD-MSC) challenge is designed around this deployment problem by evaluating performance on both seen and unseen domains. This paper presents the official baseline system and evaluation pipeline as a simple, fully reproducible reference for the CD-MSC challenge task. The baseline uses log-mel features and a multitemporal resolution convolutional neural network (MTRCNN) with species and auxiliary domain outputs, together with complete training and test scripts. The baseline system performs strongly on seen domains but degrades markedly on unseen domains, showing that cross-domain generalisation, rather than within-domain recognition, is the central challenge for practical mosquito species classification from multi-source bioacoustic recordings.

Figures & tables

Dataset	Ae.aeg	Ae.alb	Cx.qui	An.gam	An.ara	An.dir	Cx.pip	An.min	An.ste	Total
D1	123	79	672	818	1820	87	66	200	200	4065
D2	0	419	0	0	0	0	66	99	200	784
D3	192	19	0	0	0	0	68	200	200	679
D4	22	0	13	0	0	40	0	51	74	200
D5	81250	18000	71371	46180	19297	0	29554	0	0	265652
Total	81587	18517	72056	46998	21117	127	29754	550	674	271380

TABLE I: Number of clips in the full development dataset. Rows correspond to domains and columns correspond to species.

Dataset	Ae.aeg	Ae.alb	Cx.qui	An.gam	An.ara	An.dir	Cx.pip	An.min	An.ste	Total
D1	1332.65	42.32	7418.64	2978.16	18322.92	748.09	132.66	424.85	381.63	31781.92
D2	0.00	1280.58	0.00	0.00	0.00	0.00	132.66	198.99	402.00	2014.23
D3	5909.57	580.46	0.00	0.00	0.00	0.00	136.68	402.00	402.00	7430.71
D4	1737.07	0.00	758.19	0.00	0.00	1775.24	0.00	2944.18	2586.10	9800.78
D5	51187.50	11340.00	44963.73	29093.40	12157.11	0.00	18619.02	0.00	0.00	167360.76
Total	60166.79	13243.36	53140.56	32071.56	30480.03	2523.33	19021.02	3970.02	3771.73	218388.40

TABLE II: Total duration in seconds in the full development dataset. Rows correspond to domains and columns correspond to species.

Dataset	Ae.aeg	Ae.alb	Cx.qui	An.gam	An.ara	An.dir	Cx.pip	An.min	An.ste	Total
D1	111	73	0	0	0	87	59	200	200	730
D2	0	0	0	0	0	0	60	0	200	260
D3	0	17	0	0	0	0	0	200	200	417
D4	20	0	12	0	0	0	0	51	0	83
D5	73297	16575	64838	42298	19005	0	26660	0	0	242673

TABLE III: Species-domain distribution of the trainval pool set.

Dataset	Ae.aeg	Ae.alb	Cx.qui	An.gam	An.ara	An.dir	Cx.pip	An.min	An.ste	Total
D1	12	6	672	818	1820	0	7	0	0	3335
D2	0	419	0	0	0	0	6	99	0	524
D3	192	2	0	0	0	0	68	0	0	262
D4	2	0	1	0	0	40	0	0	74	117
D5	7953	1425	6533	3882	292	0	2894	0	0	22979

TABLE IV: Species-domain distribution of the test set.

Checkpoint	$\mathrm{BA}_{\mathrm{seen}}$	$\mathrm{BA}_{\mathrm{unseen}}$	DSG
Best-validation	$0.8806\pm 0.0108$	$0.1751\pm 0.0197$	$0.7055\pm 0.0248$
Final	$0.8822\pm 0.0097$	$0.1704\pm 0.0180$	$0.7118\pm 0.0235$

TABLE V: Performance of the baseline checkpoints on seen-domain and unseen-domain samples from the test set of the development dataset, reported over 10 runs.

Fig. 1: Per-species balanced accuracy of the best-validation checkpoint on the test set of the development dataset. For each species, $\mathrm{BA}_{\mathrm{seen}}$ and $\mathrm{BA}_{\mathrm{unseen}}$ are shown, with error bars indicating standard deviation over 10 runs.

Fig. 2: Domain-level species balanced accuracy of the best-validation checkpoint on the test set of the development dataset, over 10 runs.

Checkpoint	Metric	Validation	Test
Best-validation	Species accuracy	$0.9056\pm 0.0076$	$0.7827\pm 0.0068$
Best-validation	Species balanced accuracy	$0.8522\pm 0.0108$	$0.5411\pm 0.0140$
Final	Species accuracy	$0.9096\pm 0.0052$	$0.7874\pm 0.0034$
Final	Species balanced accuracy	$0.8383\pm 0.0176$	$0.5380\pm 0.0152$

TABLE VI: Overall performance of the baseline checkpoints on the validation set and the test set of the development dataset, reported over 10 runs.

Bioacoustics

Bioacoustics uses sound to study animal activity, communication, biodiversity, and ecological change.

See all 64 papers

Jun 11, 2026

Decoding Insect Song: A Multitask Semisupervised Orthoptera Bioacoustic Classifier

Passive acoustic monitoring holds great promise for ecological inference, yet existing automated tools are typically narrowly trained and non-transferable. We address these limitations with PULSE, a semi-supervised, multi-task framework for Orthoptera bioacoustics, combining weakly-supervised species classification, self-supervised learning on unlabelled field audio, and knowledge distillation from a general-purpose bioacoustic model. Our domain-adapted specialist model outperforms a state-of-the-art general model across all metrics (macro F1: 0.21 vs. 0.07; AUC: 0.74 vs. 0.45; AP: 0.32 vs. 0.19), with active learning further raising F1 to 0.34 and AUC to 0.84. Beyond classification, the learned embeddings encode ecologically meaningful structure, exposed through an interactive visualisation tool for ecological discovery.

Olga Isupova, Danil Kuzin, Ella Browning +2

Leverhulme Centre for Nature Recovery, School of Geography and the Environment, University of Oxford, Oxford, UK

Audio Representation Learning Audio Understanding Classification Bioacoustics Passive Acoustic Monitoring Semi-Supervised Learning+2

Ask PDF

May 5, 2026

Ecologically-Constrained Task Arithmetic for Multi-Taxa Bioacoustic Classifiers Without Shared Data

Training data for bioacoustics is scattered across taxa, regions, and institutions. Centralizing it all is often infeasible. We show that independently fine-tuned BEATs encoders can be composed into a unified 661-species classifier via task vector arithmetic without sharing data. We find that bioacoustic task vectors are near-orthogonal (cosine 0.01-0.09). Their separation aligns closely with spectral distribution distance, a gradient consistent with the acoustic niche hypothesis. This geometry makes simple averaging optimal while sign-conflict methods reduce accuracy by one to six percentage points. Composition also creates an asymmetric gap: species-rich groups lose accuracy relative to joint training while underrepresented taxa gain, a redistribution useful for equitable biodiversity monitoring. We verify linear mode connectivity across all taxonomic pairs, demonstrate zero-shot transfer to new regions, and identify domain negation as a boundary condition where composition fails. These results enable a collaborative paradigm for bioacoustics where institutions share only task vectors to assemble multi-taxa classifiers, preserving data privacy.

Ragib Amin Nihal, Benjamin Yen, Runwu Shi +2

Systems and Control Engineering, Institute of Science Tokyo, Japan · RIKEN BDR, Japan

Task Arithmetic Bioacoustics Model Merging Audio Classification

Ask PDF

Sep 14, 2026

BioDCASE: Active Learning for Bioacoustics

Ecological monitoring increasingly relies on machine learning models, whose performance depends on the quality and quantity of labelled data. However, obtaining these labels is costly, particularly in passive acoustic monitoring, where vast amounts of data are collected but only a small proportion can feasibly be annotated. Active learning addresses this bottleneck by prioritizing which samples should be labelled. However, progress is difficult to measure, because published methods are evaluated under different models, budgets, evaluation metrics and datasets. To address this challenge, we present the 2026 Active Learning for Bioacoustics BioDCASE challenge: a systematic evaluation of sampling methods designed to identify effective AL strategies. Participant methods were evaluated across four subsets composed of terrestrial and marine data. Across ten proposed sampling methods from seven teams, the top-ranked method achieved an area under the learning curve 26.4 % higher than random sampling at the same annotation budget, averaged over four data subsets. Significant variation in performance was observed across subsets, with the top-performing submission achieving a 67.1 % gain for the HSN subset over random sampling and a gain of 8 % for the ATBFL subset. Top-ranking submissions combined multiple acquisition signals, and diversity-based selection outperformed pure uncertainty sampling. Furthermore, there is evidence that transitioning from diversity-based to uncertainty-based selection and explicitly reducing redundancy within acquisition batches improve model training. There is also initial evidence that larger acquisition batch sizes may be increasingly beneficial later in the labelling process.

Ben McEwen, Rupa Kurinchi-Vendhan, Shiqi Zhang +3

University of Amsterdam, Amsterdam, Netherlands · Massachusetts Institute of Technology, Cambridge, Massachusetts, USA · Tampere University, Finland +2

Active Learning Bioacoustics Passive Acoustic Monitoring

Ask PDF

Abstract

Bioacoustics

Explore similar work

Decoding Insect Song: A Multitask Semisupervised Orthoptera Bioacoustic Classifier

Ecologically-Constrained Task Arithmetic for Multi-Taxa Bioacoustic Classifiers Without Shared Data

BioDCASE: Active Learning for Bioacoustics