Neonatal Pain Expression Recognition Based On Deep Learning

Researches have shown that a lot of pain stimulation can bring a series of bad effects to the growth and development of neonates,so attaching great importance to the early neonatal pain management and control is of great significance to the health of the newborn growth.Automatic neonatal pain expression recognition technology can help medical staff evaluate neonatal pain automatically and effectively improve the quality and efficiency of evaluation.Therefore,this article further explores the related technologies of neonatal pain expression recognition,and studies the application of deep learning in neonatal facial detection and face pain expression recognition.The main research work includes the following points:(1)To establish a neonatal pain expression image database.A database is the basis for the research of neonatal facial research.In this paper,based on the actual requirements of the project,there are more than 10,000 expression key frames extracted from more than 500 neonatal facial videos.After professional assessment and image normalization operation,we create a neonatal pain expression database containing four expressions(quiet,cry,low-grade pain and high-grade pain).(2)Research on neonatal face detection.Because of the difference in facial features between newborns and adults,the existing face detector can not directly apply neonatal face detection.This paper focuses on face detection based on Adboost algorithm and MTCNN face detection two algorithms in face detection of newborns.By using newborn data to train and finetune the algorithm,the rate of neonatal face detection is greatly improved.The experimental results show that compared with the Adaboost algorithm,the facial detector trained by the MTCNN algorithm has a better detection effect,and the detection of the neonatal side face,angle tilt,and slight facial occlusion has been significantly improved.The results show that the algorithm has good robustness in practical applications.(3)Research on neonatal pain expression recognition.The basic principles of the Convolutional Neural Network were studied and applied to neonatal pain expression recognition.Two methods for recognizing the pain of newborns based on the CaffeNet network and the Xception network were proposed.The experimental results show that the Xception network has better recognition performance in the two networks used in this paper.The classification recognition rate for neonatal quiet,cry,low-grade pain and high-grade pain is about 76%.In the end,this paper combines the study of neonatal face detection and pain expression recognition to achieve a complete classification system for neonatal pain expression recognition,which can realize the automatic detection of neonatal face and the recognition of pain expression in the input image.