Estimation of mass and depth of buried radioactive materials using neural networks

In this paper, we investigate the use of a backpropagation neural network (BPNN) to estimate the mass and depth of buried radioactive materials, i.e., depleted uranium (DU). A Lanthanum bromide(LaBr) detector is employed to collect the data for buried targets with different mass and at different depths. Due to the sparseness and randomness of a gamma spectrum, spectral transformation methods are implemented for background normalization and feature extraction. These spectral transformations are based on various binned energy windows determined by the particle swarm optimization (PSO) approach. The transformed data will be used as the inputs to BPNN. Compared with the original spectra, principle component analysis (PCA)-transformed spectra, and uniformly partitioned spectra, the optimized spectral transformed data can provide more accurate estimates.