The snow leopard (Panthera uncia) represents one of the most endangered large felids globally, with population monitoring presenting significant challenges due to the species’ elusive nature and remote mountainous habitat. This study presents the first automated detection system specifically designed for snow leopard identification in camera trap imagery within the Kyrgyz Republic. We propose a two-stage classification pipeline combining motion detection preprocessing with a fine-tuned MobileNetV2 architecture enhanced by squeeze-and-excitation attention mechanisms. The system was trained on a curated dataset of 2,660 images and evaluated using 5-fold stratified cross-validation. The model achieved an AUC-ROC of 97.25%, Average Precision of 92.88%, and sensitivity of 99.9% (±0.2%), missing only 1 snow leopard image out of 916 across all validation folds. Threshold optimization analysis demonstrated that adjusting the decision boundary from 0.5 to 0.95 improves specificity from 64% to 88% while maintaining 99% sensitivity. The model has been deployed as a functional web application, representing a pioneering contribution to technology-assisted wildlife conservation efforts in Central Asia. This work establishes a foundation for automated biodiversity monitoring systems in the region and demonstrates the viability of transfer learning approaches for species-specific detection tasks with limited training data.