Detection and analysis of children's facial expressions with deep learning methods

Expression identification through facial images is a very popular field of study. Many studies on defining expressions have been carried out using data sets consisting of adult individuals. Studies on these groups are required for new technologies and systems to be developed for the entire society or for different age groups. Therefore, this study focused on the detection of facial expressions of young children which have very few studies in the literature. For this purpose, a data set consisting of 4002 images was prepared by using search engines. Using the pre-trained VGG16, ResNet50, DenseNet121, Inception V3 Inception, ResNetV2 and Xception models on the prepared data set, angry, disgust, happy, neutral, sad, fearful and surprised expressions were detected. To determine the best performing model, accuracy, precision, sensitivity and F1 scores of the models used in this thesis were compeared. As a result of the study, the best result in detecting expression on children's faces was obtained with the Inception V3 model with 76.3% accuracy and 0.76 F1 score. It is thought that this recent study will contribute to the studies on detecting children's facial expressions and to the development of systems designed for children.