Classification of Big Cats Using Deep Learning

BSc. Getabalew Amtate Hailemariam

Diplomová práce

Abstract:

This study employs Design Science Research (DSR) methodology to develop and evaluate deep learning methods for proper distinction of several big cat species, an important aspect of big cat's conservation efforts. A diverse dataset featuring images of multiple big cat species was employed to evaluate the performance of different convolutional neural network (CNN) models and to demonstrate the utility of transfer learning. A successful experiment was done on training a CNN model capable of extraction of knowledge from RGB color images resulting with an approximate accuracy of 97%, which signified the importance of color properties in the effectiveness of such systems. Detailed analysis revealed that a larger dataset and colorful image dataset can improve classification accuracy. Among pre-trained models, Xception and ResNet152V2 have the best performance, with Xception achieving 99% accuracy. This result shows the efficiency of leveraging pre-trained models for specialized image classification tasks. This research significantly contributes to deep learning studies through its recognition of important color-based image models and transfer learning strategies development. The finding results offer essential research information that benefits professionals and scientists actively working in this field, particularly in advancing image classification tasks for conservation purposes.

Keywords

Big Cat Classification, Deep Learning, CNN, Transfer Learning, Xception, ResNet152V2, Wildlife Conservation, Wild cats, Image Classification

Jazyk práce: angličtina

Datum vytvoření / odevzdání či podání práce: 28. 3. 2025

Obhajoba závěrečné práce

• Obhajoba proběhla 3. 6. 2025
• Vedoucí: doc. Ing. Arnošt Veselý, CSc.
• Oponent: Richard Vágner, externi

Česká zemědělská univerzita v Praze

Provozně ekonomická fakulta

Magisterský studijní program:
Informatics