Cat Breed Recognition System

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

Text RDS__Wong Yann Yiee_Fulltext.pdf Restricted to Registered users only Download (7MB)

Abstract

There are a lot of cat breeds all over the world and people like to raise cats. To help them recognise the cat breeds such as Bengal, the Cat Breed Recognition system is useful for them. Users can upload images or show the images in front of the camera to let the system predict. For the dataset, researchers select 6 desired classes from 67 different classes from the Kaggle dataset. Data splitting and data augmentation will be implemented to get the training generator and validation generator since they are useful for model building. Also, researchers build the models using deep learning which are VGG16, MobileNetV2, ResNet50V2,DenseNet121 and InceptionResNetV2 and researchers found that DenseNet121 has the best performance which is 85.58% accuracy. For the user interface, researchers implemented the Streamlit framework to do it. It can help us to make an interface including our system functions effectively. If the user wants to predict apart from the 6 classes, it will show a message ,‘Unknown’. After the system predicts the user uploaded image, the system also will come out with a message to show the cat breed and this is based on the voting strategy. In conclusion, it has some limitations which are the number of cat breed classes that need to be taught growing, the model VGG16 has encountered various technical issues, the researcher has run into difficulty with libraries like tensorflow and opencv-python and the system interface need to improve be more user-friendly.