Neonatal Pain Expression Recognition Based On The Curvelet Transform

Neonates undergo repeated pain stimulation will produce a series of physical andpsychological impact, in the worst case, will cause severe and permanent central nervous systemdamage and developmental delay. The traditional method, used to assess neonatal pain is conductedby health professionals, has many limitations. For example, using this artificial method, plenty oftime and energy are spent to cultivate many health professionals; the results of the assessment arevulnerable to the subjective factors; sometimes, health professionals cannot assess the neonatal painimmediately. So, it is significant to develop an auto and efficient system to recognize neonatal pain.Feature extraction is the key technology in the procedure of neonatal pain recognition and itdetermines the recognition rate.This paper uses Curvelet transform to extract the feature of neonatalpain expression, then uses FDA method to reduce the dimensions of feature,and finally uses sparserepresentation as a classifier. Experiments show that the proposed algorithm has a satisfiedperformance, the average correct rate can achieve to91.75%.The main contributions are as follows:(1)The paper discusses the impact of direction parameter of Curvelet transform on recognitionresults.(2)In order to analysis the relation between recognition rate and dimension, the FDA method isadopted to reduce the dimensions of feature.(3)Several methods are used to decompose the test images, such as Newton-preconditionconjugate gradient method, homotopy and orthogonal matching pursuit method.Then therecognition performance under different number of training samples and different feature dimensionis discussed after experiments.(4)The cross validation method is adopted to calculate the average recognition rate, and thecorrect identification rate and false rate of each class in each test are analyzed at last.