Marginal Fisher Analysis-based feature extraction for identification of drug and explosive concealed by body packing

In the analysis of the energy dispersive X-ray diffraction (EDXRD) spectra of drugs and explosives concealed by body packing (i.e. the internal concealment of illicit drugs), the method of feature extraction based on Marginal Fisher Analysis (MFA) is introduced to resolve the challenge from the data of high dimension, small sample size and poor signal-to-noise ratio. MFA is applied to extract features and makes full use of both the local geometric structure (in the intrinsic graph) and label information (utilized in both graphs) to seek efficient modes of discrimination. Features extracted by principal component analysis (PCA) and PCA plus linear discriminant analysis (LDA) were investigated for comparison with the features extracted by MFA. Further, in order to avoid the influence of classifiers, two kinds of classifiers (K-nearest neighbour and support vector machine) were introduced to classify the samples according to the features. It is shown that the recognition rates obtained by MFA are more accurate (averaged recognition rate > 99.4%) compared with the other candidates. This investigation has demonstrated that MFA is effective in feature extraction for the identification of drugs and explosives concealed by body packing.