EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Improved detection and classification of cryptic ‘Phylloscopus’ warblers with integrated computer vision and deep learning method

Tushar Parab, Glen Bennet Hermon, Maitreyee Bhave, Shashank Nagarale, Arnab Chattopadhyay, Sagar Rajpurkar

Journal of Wildlife Science · 2025

DOI: 10.63033/jwls.ggig1171 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Wildlife Science, 2025. Verfasst von Tushar Parab und 5 weiteren Autor:innen. Die Arbeit nennt 40 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The rapid advancement of image recognition technology, especially with the help of YOLO algorithms, has greatly improved the ability to identify and recognize different species. This research focuses on the Phylloscopus burkii cryptic species complex, distinguished by little phenotypic diversity making precise identification challenging. In order to make it easier to distinguish between members of this complex, we suggest an enhanced YOLOv5-based model that employs a part-based approach that keeps part detection and identification classification apart. Our goals included determining whether images taken in natural settings can produce statistically reliable identifications and examining the variation of image characteristics of the species within the burkii complex. A dataset of 289 annotated images was augmented to build 2,890 photographs for training purposes. After training on Amazon Elastic Cloud, the model achieved a mean average precision (mAP) of 94 and a recall of 97.3. The model demonstrated strong precision and recall metrics across species, and the results showed that it worked best at a confidence threshold of 0.50. P. valentini and P. tephrocephalus, in particular, showed considerable overlap, which sometimes resulted in misidentifications, although P. poliogenys and P. whistleri demonstrated excellent accuracy. We recommend training models based on species geographic distributions to improve identification accuracy. This strategy might successfully lower the rate of erroneous identifications, enhancing the caliber of data produced by citizen science projects. In the end, our results support current efforts to conserve biodiversity by offering a solid framework for automating species identification and guiding ecological study. Our goal is to improve species monitoring and data collection precision by utilizing innovative methods based on machine learning.

Zitieren

APA

Parab, T., Hermon, G. B., Bhave, M., Nagarale, S., Chattopadhyay, A., & Rajpurkar, S. (2025). Improved detection and classification of cryptic ‘Phylloscopus’ warblers with integrated computer vision and deep learning method. Journal of Wildlife Science. https://doi.org/10.63033/jwls.ggig1171

Vancouver

Parab T, Hermon GB, Bhave M, Nagarale S, Chattopadhyay A, Rajpurkar S. Improved detection and classification of cryptic ‘Phylloscopus’ warblers with integrated computer vision and deep learning method. Journal of Wildlife Science. 2025. doi:10.63033/jwls.ggig1171

BibTeX
@article{Parab2025,
  author = {Parab, Tushar and Hermon, Glen Bennet and Bhave, Maitreyee and Nagarale, Shashank and Chattopadhyay, Arnab and Rajpurkar, Sagar},
  title = {Improved detection and classification of cryptic ‘Phylloscopus’ warblers with integrated computer vision and deep learning method},
  journal = {Journal of Wildlife Science},
  year = {2025},
  doi = {10.63033/jwls.ggig1171}
}
Kostenlos starten →