Research on XRF Spectral Classification of Bumpers Based on Particle Swarm Optimization BP Neural Network
A rapid non-destructive testing method for bumper analysis has been established.50 bumper samples were measured for element types and contents using a handheld X-ray fluorescence spectrometer.Based on the relationship between K-means algorithm and contour coefficient,sum of squared errors within clusters(SSE),the optimal number of clusters was determined to be 5,which means that the 50 samples were divided into 5 categories using K-means cluste-ring algorithm.By using the random forest(RF)algorithm,features are extracted from the X-ray fluorescence(XRF)spectral data of samples.Based on the different feature variable combinations extracted by the RF algorithm,a back propagation neural network(BPNN)and a particle swarm optimization(PSO)optimized BPNN(PSO-BPNN)were es-tablished.The results show that when the input variables are Ca-Pb-Sr element variables,both BPNN and PSO-BPNN have good classification performance,with classification accuracies of 94％and 98％,respectively.The PSO-BPNN model is more suitable for XRF spectral data of such samples.The combination of XRF and PSO-BPNN can achieve ef-fective classification of bumpers.This method is simple,fast,and non-destructive for samples.It can provide scientific basis for the identification of bumper physical evidence.
万方数据
维普
