This research work introduces a novel method for animal recognition by leveraging transfer learning with MobileNetV2 as the base model. Addressing challenges in accurate animal classification, our study investigates the strategic fine-tuning of upper layers to enhance model generalization. The distinctive contribution lies in extending transfer learning to animal detection, specifically targeting perspective challenges. The key novelty is encapsulated in a meticulously crafted custom dataset, bolstered by data augmentation techniques. This dataset diversity equips the model to tackle real-world complexities effectively. Experimentally the method was evaluated culminating in remarkable accuracy for binary classification, particularly in distinguishing cat images. Our findings underline the efficacy of this innovation, illuminating a path toward robust animal detection. As technology and ecological preservation intertwine, our research not only advances methodology but also extends its implications to biodiversity conservation. In a landscape of evolving challenges, our work showcases the potency of transfer learning and MobileNetV2, opening vistas for nuanced animal detection paradigms. After fine-tuning the model, the highest accuracy of 96.75% was obtained without overfitting and the model attained better generalization capability. This model can be easily deployed on embedded systems because of its smaller size.