Blood identification of NIR spectroscopy based on BP neural network combined with particle swarm optimization

Since the disadvantages of destructive and high cost for the traditional detection methods, a non-invasive identification method based on near infrared spectroscopy was used in this paper. The near infrared spectra of 120 groups of training blood samples and 30 groups of test samples were obtained from 4000cm-1 to 10000cm-1 for four kinds of animal blood and two kinds of fake blood. The classification and identification of blood can’t be easily achieved because of the near infrared spectra overlapping. To accurately identify the different kinds of the blood, back propagation (BP) neural network was used to establish the classification model. The spectra in full wavelengths were used as the input data, and 1, 2, 3, 4, 5, 6 were used to label different blood. Based on the training of 120 groups of training blood samples, the correct rate of blood identification for 30 groups of test samples are 66.7%. To further improve the correct rate, the weights of BP neural network were optimized by the particle swarm optimization (PSO). The effects of neurons number, learn rate factor, iteration times, and training times on the correct rate and mean square error for the identification of blood based on BP-PSO algorithm were investigated. Under the optimized parameters, the correct rate was improved to 96.7%.