Recognition of Fingernail Region Based on Filter£­Spectral Feature Extraction

KeyWord:spectroscopy  ?fingernail  ?Hilbert transform filter  ?principal component analysis£¨PCA£©  ?machine learning

Abstract:

      Fingernail is one of the common biological evidences on the scene of the case£®The rapid inspection of fingernails found in the scene could provide direction and clues for case investigation£®Meanwhile£¬application of machine learning for quick and nondestructive detection of the testing material is an important branch of court science£®Filter could effectively remove the noise and background interference of the spectra£®The dimension reduction of the spectral data could effectively reduce the dimension of the data£¬and improve the recognition effect of the model£®In this paper£¬a total of 204 nail samples from the actual cases of five regions were collected£®The original spectra were denoised by Hilbert transform filter£¨HTF£©£¬and then the principal component analysis£¨PCA£©was used to reduce the dimension of the original data and the denoised data£®Naive Bayes£¨NB£©£¬random forest£¨RF£© and partial least squares discriminant analysis£¨PLS£­DA£© model were used to carry out the identification of nail area£®According to the recognition rate and related indicators of the model£¬the optimal preprocessing method and optimal recognition model for nail area identification were selected£®The results demonstrated that the recognition rate of the original spectra is significantly improved after preprocessing£®HTF combined with PCA is the best preprocessing method£®The recognition rate of RF for the training set of the best pretreatment method is 94.88%£¬while that for the test set is 93.47%£®This method could effectively reduce the noise of spectra£¬reduce the redundancy of data£¬improve the recognition effect of the model£¬and provide some reference for the rapid identification of nail areas in forensic science£®