Investigating the emotional lives of animals is inherently complicated although the findings are rewarding in terms of conservation, psychology, and sociopsychology. In this paper, we consider the problem of animal emotion detection with the aid of deep learning using dogs’ emotions in this study. In this case, four categories of emotions, angrily, joyfully, relaxed, and sadly were classified based on the best CNNs such as ResNet-50, EfficientNet, and MobileNet. Furthermore, by employing data preprocessing, data augmentation techniques, and structural explainability techniques like the Grad-CAM, we could improve how the model made critical decisions. Results confirmed that the model performed satisfactorily in detecting the subjects’ emotions even though joyful and relaxed states were more pronounced with high levels of accuracy compared to others with emotions like sadness and anger lauded as a notable challenge, especially in discriminating attitudes that seemed too close to one another. The application of Grad-CAM was able to elaborate on the regions of interest incorporated by the model thus enhancing the explainability of the model. This paper is focused on the emergent developments in emotion detection in animals and further recommends for advancement of three-dimensional deep learning techniques, settlement of the dataset, and the introduction of more complicated explainable AI techniques such as Local Interpretable Model Explanations (LIME).