Building of a Convolutional Neuronal Network for the prediction of mood states through face recognition based on object detection with YOLOV8 and Python

This research focuses on developing a real-time mood state prediction CNN model from the extraction of facial expressions. It targets 6 mood states (Tension, Frustration, Anxiety, Fatigue, Neutrality, Happiness) via facial feature extraction and pattern recognition. . The project begins with the creation of the dataset which is later preprocessed to standardize image resolutions and lightning conditions to generate additional training examples to augmente the images from 2500 to 4000 images for training. Adopting the agile incremental development approach to train the CNN, it is ensured that each iteration of the model is fully functional and builds upon the model’s previous versions, using them as checkpoints, to focus on the maximization of the precision and mAP value, maintaining a mAP value of 99.5% in each iteration with little to no fluctuation on the 99% precision value on every version of the model. The research then transitions to manual training using the YOLOv8 object detection framework on google colab jupyter notebook service. The YOLOv8 model achieve consistent high precision during its 155 epoch training, obtaining the model’s best weight at epoch 129, with a precision of 94.3%. This results indicates the potential of the model for the development of systems that responds to human feelings effectively.