Bridging AI and wildlife conservation: Classifying wildcats from genetically validated images
Year: 2025
Type: article
Abstract: This study presents a multidisciplinary approach for classifying three cat categories, European wildcats (Felis silvestris, WC), hybrids (F. silvestris × catus, HY), and domestic cats (F. catus, DC), using convolutional neural networks (CNNs) on real-world images collected via camera traps, roadkills, and research-team sources. The initial dataset included 3,392 images(WC = 1,738; HY = 461; DC = 1,193) and was filtered using YOLOv12 to retain clearly visible cats, resulting in 2,518 images (WC = 1,201; HY = 461; DC = 856). Among the five evaluated CNN architectures, MobileViT-V2 achieved the highest accuracy (86%) on the balanced dataset. Tested on a genetically validated dataset, it achieved 71% classification accuracy. When trained on the larger, unbalanced dataset, accuracy on the genetic test increased to 81%, suggesting that more data may compensate for class imbalance. An offline-capable mobile app was developed for field use, supporting local image caching and usability feedback via an included and structured questionnaire for future analysis. This work results from a synergistic collaboration among computer scientists, and networks of wildcat researchers and citizen scientists. The proposed model offers a robust tool for supporting European wildcat conservation through real-world classification.
Source: Ecological Informatics
Institutions: University of Catania, University of Palermo
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,770
Award: G980899861
OpenAlex ID: W4415320641