Accurate prediction of copper content in low-grade copper ore using PGNAA and an ACNN algorithm

To predict the copper content of low-grade copper ore in blast holes, this study proposes an Attention Convolutional Neural Network (ACNN) model by integrating a Convolutional Neural Network (CNN) with an attention mechanism. First, based on Prompt Gamma Neutron Activation Analysis (PGNAA), a copper-ore measurement system equipped with 252Cf source and a bismuth germanate (BGO) detector was established.

The gamma spectra were obtained by measuring laboratory-prepared copper-ore samples with this system. Subsequently, the Discrete Cosine Transform (DCT) was applied to the spectral data for feature extraction. The hyperparameters of the ACNN model were optimized using Bayesian optimization.