An advanced deep learning approach for cat breed classification

Abstract

In this paper, an improved deep learning method is proposed to recognize the cat breeds with high accuracy and high CNN structures. The dataset, sourced from Kaggle, comprises 2,557 labeled images across four distinct cat breeds: These breeds include Ginger Cat, Bombay Cat, Bengal and Sphynx Cat. The primary objective was to evaluate and compare the performance of five deep learning models: Xception, CNN, VGG19, MobileNetV2, and InceptionResNetV2. Comparing the efficiency of all models the MobileNetV2 model reached the highest accuracy reaching the mark of 99.87%. This efficient architecture appropriate for mobile and embedded platforms outperformed other algorithms of its kind in feature extraction and learning engineer fine features necessary to differentiate between cat’s breeds based on their visual cues. This was followed by data preprocessing steps such as normalization and augmentation, to improve variability of the developed datasets and model. The use of transfer learning from pre-trained models, especially from ImageNet, also aided in faster convergence and superior performance across all the models under analysis. The presented results demonstrate the effectiveness of the most sophisticated deep learning methods in attaining a high level of conclusive accuracy to classify the cat breeds and their appearances accurately. The findings of this study provide necessary and useful knowledge for practicing veterinarians, authorities, pet shelters, agencies, and others in the use of AI techniques in the processing of veterinary diagnostics, care, pets’ adoption, and shelters. Future research could investigate pooling techniques and the use of higher and a more diverse amount of data to increase accuracy and the scope of the classifier’s utilization.