Quality-driven unsupervised data curation and robust learning method for bird image data

Abstract

Data acquisition for river-lake avian species suffers from interference by long-distance imaging, water surface reflections and occlusions, producing low-quality images with motion blur, low contrast and annotation noise. Such defects severely degrade the accuracy of classification models. To address this problem, this paper proposes a processing algorithm for low-quality river-lake bird image data, termed the quality-driven unsupervised data curation and robust learning (QUC-RL) Method. The algorithm is divided into four key modules: First, a multi-dimensional quality-aware representation space for bird image data is constructed by fusing deep semantic features, texture features and four interpretable quality metrics, based on which a quality score is derived. Second, multi-strategy unsupervised auditing and curation for outlier and mislabelled sample localization is implemented with the constructed multi-strategy cluster selection mechanism. Third, the dataset is preliminarily reconstructed via an adaptive distribution preservation strategy based on the obtained quality scores and localization results. Finally, hard subsets within the reconstructed dataset are enhanced by a quality-conditional robust learning framework. Experimental results demonstrate that the overall performance is preserved and the stability of macro-average performance on low-quality subsets is improved, while the dataset scale is reduced by the proposed QUC-RL method. Classification accuracies of 96.580%, a Macro-F1 of 95.807% and a Grade C F1 of 96.362% are achieved by models trained on the reconstructed dataset.

Keywords

Low-quality image recognition / deep learning / unsupervised data clustering / robust learning

References

[1] Prychepa M.Waterfowl as indicators of the state of wetland ecosystems. In Innovation in Science: Global Trends and Regional Aspect, Proceedings of the Conference, Riga, Latvia, March 12-13, 2021; Baltija Publishing: Riga, 2021; pp. 9-12.

[2] Peng J,Liao X.Wild animal survey using UAS imagery and deep learning: modified Faster R-CNN for kiang detection in Tibetan Plateau.ISPRS J Photogramm Remote Sens2020;169:364-76

[3] Chen X,He Y.An efficient method for monitoring birds based on object detection and multi-object tracking networks.Animals2023;13:1713 PMCID:PMC10215751

[4] Zhang Z,Lu B.Temporal insights into ecological community: advancing waterbird monitoring with dome camera and deep learning.J Environ Manage2025;387:125769

[5] Wu E,Lu H.Unlocking the potential of deep learning for migratory waterbirds monitoring using surveillance video.Remote Sens2022;14:514

[6] Chalmers C,Wich S.Removing human bottlenecks in bird classification using camera trap images and deep learning.Remote Sens2023;15:2638

[7] Mulero-Pérez D,Ramirez-Gordillo T.A federated learning architecture for bird species classification in wetlands.J Sens Actuator Netw2025;14:71

[8] Oba Y.Accelerating ecosystem monitoring through computer vision with deep metric learning.Ecol Complex2025;62:101124

[9] Wah C,Welinder P,Belongie S. The Caltech-UCSD Birds-200-2011 Dataset. 2011. https://authors.library.caltech.edu/records/cvm3y-5hh21. (accessed 2026-06-24)

[10] Ma J,Zheng X.An improved bird detection method using surveillance videos from Poyang Lake based on YOLOv8.Animals2024;14:3353 PMCID:PMC11639462

[11] Fang S,Zou H,Jin W.Birds-YOLO: a bird detection model for Dongting Lake based on modified YOLOv11.Biology2025;14:1515 PMCID:PMC12650164

[12] Huang Q,Hu C,Wang Y.Waterbird image recognition using lightweight deep learning in wetland environment.Avian Res2025;16:100306

[13] He J,Liu S.TransFG: a transformer architecture for fine-grained recognition.AAAI2022;36:852-60

[14] Du R,Bhunia AK. Fine-grained visual classification via progressive multi-granularity training of jigsaw patches. arXiv 2020, arXiv:2003.03836. Available online: https://doi.org/10.48550/arXiv.2003.03836. (accessed 2026-06-24)

[15] Xie S,Liu Y.Step-by-step to success: multi-stage learning driven robust audiovisual fusion network for fine-grained bird species classification.Avian Res2025;16:100280

[16] Zha D,Lai KH. Data-centric artificial intelligence: a survey. arXiv 2023, arXiv:2303.10158. Available online: https://doi.org/10.48550/arXiv.2303.10158. (accessed 2026-06-24)

[17] Song H,Park D,Lee JG.Learning from noisy labels with deep neural networks: a survey.IEEE Trans Neural Netw Learn Syst2023;34:8135-53

[18] Northcutt C,Chuang I.Confident learning: estimating uncertainty in dataset labels.JAIR2021;70:1373-411

[19] Feuer B,Cohen N. SELECT: a large-scale benchmark of data curation strategies for image classification. arXiv 2024, arXiv:2410.05057. Available online: https://doi.org/10.48550/arXiv.2410.05057. (accessed 2026-06-24)

[20] Drenkow N.A causal framework for aligning image quality metrics and deep neural network robustness.npj Artif Intell2025;1:24

[21] Liu Y,Che X,Lu G.Deep learning based fine‐grained image classification: recent advances, applications and future outlook.IET Image Process2025;19:e70243

[22] Zhang L,Gao F.Q-Norm: robust representation learning via quality-adaptive normalization. In 2025 IEEE/CVF International Conference on Computer Vision (ICCV), Honolulu, USA, October 19-23, 2025; IEEE: 2025; pp. 13901-11.

[23] Jiang K,Liu X,Lin CW.PH-Mamba: enhancing Mamba with position encoding and harmonized attention for image deraining and beyond.IEEE Trans Image Process2026;35:1727-39

[24] Xiao Y,Jiang K,Wang S.Multi-axis feature diversity enhancement for remote sensing video super-resolution.IEEE Trans Image Process2025;34:1766-78

[25] Jiang K,Yi P.Rain-free and residue hand-in-hand: a progressive coupled network for real-time image deraining.IEEE Trans Image Process2021;30:7404-18

[26] Müller SG. TrivialAugment: tuning-free yet state-of-the-art data augmentation. arXiv 2021, arXiv:2103.10158. Available online: https://doi.org/10.48550/arXiv.2103.10158. (accessed 2026-06-24)

[27] Wang P,Wen H.LLM-AutoDA: large language model-driven automatic data augmentation for long-tailed problems. In Advances in Neural Information Processing Systems 37 (NeurIPS 2024), Vancouver, Canada, Dec 10-15, 2024; Curran Associates, Inc.: 2024; pp. 115783-814.

[28] Pearson K.LIII. On lines and planes of closest fit to systems of points in space.Lond Edinb Dublin Philos Mag J Sci1901;2:559-72

[29] Pech-Pacheco JL,Chamorro-Martinez J.Diatom autofocusing in brightfield microscopy: a comparative study. In Proceedings of the 15th International Conference on Pattern Recognition, Barcelona, Spain, September 3-7, 2000; IEEE: 2000; pp. 3318-21.

[30] Moulden B,Gatley LF.The standard deviation of luminance as a metric for contrast in random-dot images.Perception1990;19:79-101

[31] Duda R. Pattern classification and scene analysis. Wiley: 1973. https://books.google.com/books?id=POMGRAAACAAJ&source=gbs_ViewAPI. (accessed 2026-06-24)

[32] Hampel FR.The influence curve and its role in robust estimation.J Am Stat Assoc1974;69:383-93

[33] Bao Y,Yang L,Zhao B. Normalizing batch normalization for long-tailed recognition. arXiv 2025, arXiv:2501.03122. Available online: https://doi.org/10.48550/arXiv.2501.03122. (accessed 2026-06-24)

[34] Mcinnes L,Astels S.hdbscan: hierarchical density based clustering.J Open Source Softw2017;2:205

[35] Gagolewski M,Cena A.Are cluster validity measures (in) valid?.Inform Sci2021;581:620-36

[36] Gagolewski M.A framework for benchmarking clustering algorithms.SoftwareX2022;20:101270

[37] Zhang H,Dauphin YN. mixup: beyond empirical risk minimization. arXiv 2017, arXiv:1710.09412. Available online: https://doi.org/10.48550/arXiv.1710.09412. (accessed 2026-06-24)

[38] Murphy KP. Probabilistic machine learning: an introduction. MIT Press: 2022. https://probml.github.io/pml-book/book1.html. (accessed 2026-06-24)

[39] Liu W,Erhan D.SSD: single shot MultiBox detector. In: Leibe B, Matas J, Sebe N, Welling M, editors. Computer Vision - ECCV 2016. Cham: Springer International Publishing; 2016. pp. 21-37.

[40] Qin D,Delakis M.MobileNetV4: universal models for the mobile ecosystem. In: Leonardis A, Ricci E, Roth S, Russakovsky O, Sattler T, Varol G, editors. Computer Vision - ECCV 2024. Cham: Springer Nature Switzerland; 2025. pp. 78-96.

[41] Wang J,Zhou X.Joint asymmetric loss for learning with noisy labels. In Proceedings of the IEEE/CVF International Conference on Computer Vision, Honolulu, USA, October 19-23, 2025; pp. 1947-56.

[42] Woo S,Hu R.ConvNeXt V2: co-designing and scaling ConvNets with masked autoencoders. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR); 2023; pp. 16133-42.