Blood discrimination based on NIR spectroscopy and BP neural network combined with genetic algorithm

Abstract

Blood distinction has important role in the fields of biomedical diagnosis, animal quarantine, criminal investigation and blood products safety. The non-invasive detection will be the inevitable trend in the future development of blood distinction. In the work, the near infrared spectroscopy method was used to distinguish the real blood and fake blood. In the experiments, four kinds of animal blood (horse, cow, rabbit and sheep) and two kinds of fake blood (props blood and red ink) in total of 150 groups were used. And the near infrared (NIR) spectra of all blood samples were captured from 4000cm^-1 to 10000cm^-1 by using a set of Fourier transform NIR spectroscope in the diffuse reflection acquisition mode. Due to the problems of serious spectra overlap, the accuracy distinction of real blood and fake blood is very difficult between the real blood or between the fake blood although the spectra difference is large between the real blood and the fake blood. To ensure the accuracy of distinguishing the real and fake blood, back-propagation (BP) neural network algorithm was used. The NIR spectra of all blood samples with full wavelengths were used as the input data, and 1, 2, 3, 4, 5 and 6 were used to label the different kinds of blood. After training of 120 groups of training blood samples, 30 groups of test blood samples were used to test the accuracy rate of distinction of blood. The correct rate is 66.7%. To improve the correct rate, the genetic algorithm (GA) was used to optimize the weights and thresholds of BP neural network. Moreover, the effects of the number of neurons in the hidden layer, the learning rate factor, iteration times, and the training times on the correct rate and the mean square error were all investigated. Under the optimized parameters, the correct rate of the BP-GA algorithm can reach 96.7%.