Study on the detection of heavy metal lead (Pb) in mussels based on near-infrared spectroscopy technology and a REELM classifier

Abstract

It is toxic to consume mussels polluted by heavy metal Pb for humans. A quick, accurate, and non-destructive method based on near-infrared reflection spectroscopy (NIRS) has been investigated for the detection of Pb-polluted mussels in this paper. The spectral data of non-polluted and Pb-polluted mussels in the range of 950–1700 nm is acquired. A wavelength selection algorithm based on information measures of neighborhood rough set is applied to select optimal wavelengths. A sparse extreme learning machine (ELM) based on residual errors (REELM) is established as a classifier to detect Pb-polluted mussels. The average accuracy of 50 randomly assigned test datasets reaches 99.27% for detecting Pb-polluted mussels. For the class imbalance problem and mislabeled samples in the presence of practical applications, the detection performance of the proposed model has been analysed. The experimental results show that, compared with the traditional ELM and randomly pruned ELM, the REELM model is proven to be superior. The results indicate that NIRS combined with pattern recognition method has great potential to detect Pb pollution in mussels. This research is of significant importance in terms of the evaluation of edible quality and safety of mussels.