Recognition For Facial Expression Of Pain In Neonates Based On Convolutional Neural Network

Neonatal pain expression recognition is a special application of facial expression recognition,The study looked at the facial expressions of newborn babies in pain.The traditional machine learning method relies on artificial feature design and feature extraction to study facial expression recognition.It not only does not fully characterize the essence of facial expression,but also has some complexity in the process of implementation.In recent years,the depth learning method,represented by Convolutional Neural Network(CNN),has made great progress in image analysis,especially in image classification.Therefore,in this paper,the convolution neural network algorithm is introduced into the algorithm expression recognition of neonatal pain in order to realize the automatic evaluation of neonatal pain.The main research work and related achievements are summarized as follows:1)Study the newborn face detection based on Adaboost algorithm.Study on using the Haar feature of Adaboost algorithm in the basic theory and experiment in newborn face detection,Generate cascade classifiers through experiments of iterative learning,to achieve the correct face detection of neonatal success rate of 76.8%,completed by neonatal face detection 2)in the process of the establishment of neonatal pain facial image database,and 4)in neonatal pain facial expression recognition system provides a face detection function.2)To establish the video database of neonatal pain expression and the database of neonatal pain expression image.The video data of newborns in different states were collected in hospital pediatrics,and a video database of 509 samples of neonatal pain expression was established according to the scientific evaluation classification of medical staff.At the same time,According to the video database,a database of 11000 samples of neonatal pain expression images was established.The video database and image database included four basic states of calm,crying,mild pain and severe pain.3)To study the recognition of neonatal pain expression based on CNN algorithm.This paper summarizes the basic theory and network structure of CNN algorithm,By using CNN algorithm to carry out multi-core convolution and pool sampling to realize the mapping of pain expression image from lower features to high-level features,and using Softmax classifier to classify features.Through several experiments on VGGNet network,AlexNet network and AlexNet network simplified:recognition of calm and non calm state rate of 98.72% in the neonatal pain facial image database;recognition of pain and non pain two classification rate of 82.80%;three classification(calm /crying / pain)recognition rate of 89.85%;four(classification calm / cry / mild pain / severe pain)recognition rate of 71.27%.On the same data set,compared with LBP+SVM algorithm,the recognition rate of CNN algorithm increased by 4.51%,-2.3%,8.45% and 5.31%,respectively.Compared with HOG+SVM algorithm,the recognition rate of CNN algorithm increased by 3.12%,-3.31%,6.35% and 2.41%,respectively.Compared with LBP+HOG+SVM algorithm,the recognition rate of CNN algorithm increased by 2.62%,-4.6% %,5.15% and 1.51%,respectively.4)Implement the expression recognition system of neonatal pain based on CNN algorithm.On the basis of integrating the research results of 1)and 3),a new neonatal pain expression recognition system based on CNN algorithm is designed.The system can detect the state of neonatal pain expression in the input video.The evaluation results are output in real time in the system.