Cat Breed Recognition System

Yap, Johnston (2024) Cat Breed Recognition System. Final Year Project (Bachelor), Tunku Abdul Rahman University of Management and Technology.

Text RDS_Johnston Yap_FullText.pdf Restricted to Registered users only Download (16MB)

Abstract

The popularity of cat ownership has spurred the development of tools and technologies to assist cat enthusiasts in identifying various cat breeds. In response to this demand, we present a Cat Breed Recognition system designed to accurately identify 6 desired cat breeds from a diverse dataset of 67 different feline classes sourced from Kaggle. Leveraging deep learning techniques, our system enables users to upload images or use their camera to capture images for real-time prediction. Data preprocessing, including data splitting and data augmentation, is a pivotal aspect of our system. These techniques are employed to create training and validation data generators, enhancing the effectiveness of our model building process. We explore the capabilities of five well-known deep learning architectures: VGG16, MobileNetV2, ResNet50V2, DenseNet121, and InceptionResNetV2. Through rigorous evaluation, we identify DenseNet121 as the top-performing model, achieving an impressive accuracy of 85.58%. To facilitate user interaction, we have developed a user-friendly interface using the Streamlit framework in an localhost usecase. This interface seamlessly integrates our system's functionalities, allows to effortlessly upload images or capture them using their camera. In cases where the system encounters an image depicting a cat breed outside the six pre-defined classes, it will provide a response of 'Unknown.' The core feature of our system is its ability to predict the cat breed accurately and efficiently. To enhance prediction reliability, we employ a voting strategy, ensuring that the final result is based on a consensus among the deep learning models.