Dog Emotion Recognition Using Deep Learning with an Expandable Image Dataset

This work presents a deep learning-based system for recognizing canine emotions using facial images. The proposed model integrates a MobileNetV2 backbone with an LSTM module to capture spatial dependencies in image features. An expandable dataset structure is introduced, allowing continuous improvement without code modification. The system achieves 72.5% test accuracy and is deployed as a real-time Streamlit web application. The approach demonstrates improved performance over CNN and CNN+RNN baselines for three emotion classes: Angry, Happy, and Sad.