A NASNetMobile-Based Deep Learning Framework for Cat and Dog Recognition

Abstract

Animal recognition has emerged as a non-invasive approach that is increasingly adopted in various applications, including smart farming and image-based animal identification. This study proposes a Convolutional Neural Network (CNN)-based animal recognition system for classifying two companion animal species, namely cats and dogs. The experiments were conducted using the Kashtanka Pets (Animal Recognition) dataset, which consists of 400 images, including 200 cat images and 200 dog images. Prior to model training, the dataset underwent several preprocessing steps, including image resizing, pixel value normalization, contrast enhancement, and image augmentation to improve data diversity and reduce the risk of overfitting. The proposed model employs a transfer learning approach using the NASNetMobile architecture as the feature extractor, followed by a customized classification head for binary classification. The network was trained using the Adam optimizer with the categorical cross-entropy loss function and evaluated using accuracy, confusion matrix, precision, recall, and F1-score. Experimental results demonstrate that the proposed model achieved an overall test accuracy of 70%, while maintaining relatively balanced precision, recall, and F1-score across both classes. The confusion matrix analysis further indicates that misclassification errors were distributed relatively evenly between the cat and dog classes.The findings suggest that the transfer learning-based CNN is capable of effectively extracting discriminative visual features for animal recognition, although there remains considerable room for improving classification performance. Therefore, the proposed approach provides a promising foundation for the development of more robust image-based multi-species animal recognition systems.