Distinction of true and fake blood based on near infrared spectroscopy and wavelet neural networks

The accurate distinction of the true and fake blood is very important work in the fields of biomedical treatment, criminal investigation, animal inspection, food safety, etc. In this work, the near infrared spectroscopy was used to obtain the spectral of blood samples. Four kinds of animal blood (horse, cow, rabbit, and sheep) and two kinds of fake blood (props blood, and red ink) were test. 120 groups and 30 groups of blood were used as the training sample and test samples. In the experiments, the optical diffuse reflection spectra of blood samples were obtained at the wave-number from 4000cm-1 to 10000cm-1. From the experimental results, it can be seen that the optical spectra profiles are very similar between the animal blood and between the fake blood, although the spectra profiles between the true blood and the fake blood are different. At the same time, the spectra overlap are also existed in the true and fake blood, which results in the difficulty of distinguishing the different types of blood. To rapidly and accurately distinguish the true and fake blood, the wavelet neural networks (WNN) algorithm was used to train the training blood samples. Under two optimal learning rate factors, the correct rate of distinguishing true and fake blood is about 23.3%. To improve the correct rate, the genetic algorithm (GA) was used to optimize the weights, thresholds, scaling and translation factors of WNN. Under the optimal parameters of WNN-GA algorithm, the correct rate reaches 56.7%. To further improve the correct rate, the principal components analysis (PCA) algorithm was combined into WNN-GA, i.e., PCA-WNN-GA algorithm was used. The correct rate can reach 96.7% under the optimal principal components. Therefore, near infrared spectroscopy combined with WNN-GA algorithm has the potential value in the identification of blood origin.