Comparison and applicability of deep learning models for emotion recognition

In this thesis, a study was conducted in the field of emotion detection by utilizing deep learning architectures to identify and classify various facial expressions. The aim of this study is to distinguish between different facial expressions such as anger, contempt, disgust, fear, happiness, neutrality, sadness, and surprise. The study includes the reasons for using well-known models with different features and capabilities in the field of computer vision and machine learning, such as DenseNet121, EfficientNetB5, MobileNetV2, ResNet50, VGG16, and YOLOv8m-cls, along with their comparative results. Throughout the research, datasets containing various emotional expressions were meticulously prepared to provide comprehensive model training and evaluation. The datasets were pre-processed to enhance the feature extraction process and reduce potential biases. Subsequently, each model was trained using necessary methods to optimize their performance in accurately detecting and classifying facial expressions. The results exposed the accuracy and weaknesses of each architecture, shedding light on their applicability in real-world scenarios. The findings highlight the importance of selecting models that are suitable for specific use cases and computational constraints. In the future, this research has the potential to support further investigation and development of emotion detection techniques, providing benefits in various fields such as healthcare, human-computer interaction, and affective computing. Among the given models, the YoloV8m-cls model achieved the highest accuracy with 89%, while the VGG16 model showed the worst performance and the largest losses with an accuracy of 60%, making it an unsuitable method.