Patient's Pain Recognition by Using Deep Models Based on Transfer Learning

Abstract

Undetected pain threatens the quality of human life by prevents them from understanding the nature of their pain. Therefore, an urgent need to find an alternative that can recognize pain in patients suffering from difficulty in express their pain for reasons related to age or the ability of explain feeling of pain. The structure of a proposed system depends on the transfer learning technique, where two pre-trained deep convolution neural network (ResNet-50and Xception) models, have been used as a feature extractor. These models receive the images from the database, which was collected especially for this work from 100 subjects of (10–55) ages in different times and natural environments. Extracted features, which formed a features vector, were taken from the Global Average Pooling (GAP) layer. These feature vectors are passed to a supervised classification method such as Support Vector Machine (SVM), K-Nearest Neighbour (KNN), and build a Deep Neural Network (DNN) classifier from scratch. The proposed model's performance was evaluated by comparing it with related work, which adopted the concept of a pre-trained model and used the database collected in a natural environment in the training and testing process. The proposed model demonstrates outstanding results by the combination of exploiting the Transfer Learning (TL) technique in the Xception model and building the DNN as a classifier. The accuracy of the proposed model was 98.17% which demonstrates promising results.