Neonatal Pain Expression Recognition Based On The HLACLF

At present assessing neonatal pain mostly depends on the observation of doctors or nurses.These methods will cost a great quantity of time and effort. And also will be affected by subjective feeling of observers. So, it is significant to develop an automatic system to recognize neonatal pain through neonatal expression efficiently and accurately.The most important aspect of a pain expression recognition method is extraction, and it will determine recognition results. This paper selects HLACLF as the method to extract the feature of neonatal pain expression, then uses mRMR method to choose the valid feature, and finally utilizes SVM method to classify the features.The research process of this paper is as follows:(1) This paper discusses HLAC algorithm and proposes HLACLF-based feature extractionalgorithm.The effects of HLACLF with different orders (N=2or N=3) and windows of different sizes (3×3,5×5,7×7) need to be considered. And this paper overcomes HLACLF’s disadvantage onscale-invariance.(2) By calculating correlation and redundancy of feature, mRMR method selects the best features to achieve low dimensionality. The relationship between the number of feature after selection andthe results is analyzed.(3)Using traditional SVM classifier to construct Multiple-class SVM classifier has four methods:1-a-1,1-a-r, tree-SVM classifier, cross combination SVM classifier.With the algorithm proposed in this paper, experiments of the neonatal pain expression achievean average correct rate of83.75%and demonstrate the flexibility and effectiveness of thesealgorithm.