Discussion on dual-tree complex wavelet transform and generalized regression neural network based concentration-resolved fluorescence spectroscopy for oil identification
There has been a growing concern in recent years about increasing occurrence of spilled oils to the environment and proven toxic potential of these pollutants on human health and wildlife. Precisely and rapidly determining the sources of spilled oils can provide scientific evidence for the investigation and handling of spilled oils accidents. Despite traditional fluorescence spectroscopy detects in linear concentration range, a concentration-resolved fluorescence spectroscopy (CRFS) is proposed in this paper, which introduces concentration as a new dimension. A data processing strategy combining multiple algorithms is applied to the CRFS for oil spill identification. Dual-tree complex wavelet transform (DTCWT) is used to extract multi-scale and multi-directional features of CRFS to ensure the accuracy of identification, while principal component analysis (PCA) to reduce the dimension of the feature spectrum for the purpose of improving the identification speed. Three kinds of artificial neural networks (back propagation neural network BP, probabilistic neural network PNN, and generalized regression neural network GRNN), which are used as powerful classifiers for oil identification, are compared based on the spectral data processed by DTCWT and PCA. With 100% accuracy, GRNN is proved to be more suitable for oil classification and identification, especially for the case of small sample size. The combination of the CRFS technique and this data processing strategy reveals as a powerful methodology to differentiate a challenging sample set involving diesel (diesel 2002), fuel (heavy fuel 4#) and crude oils (Xia, Shang, Zhengqi), offering potential application for real-time and economic oil fingerprint identification.