Eating mussels contaminated with the heavy metal lead is a serious hazard to the human body.A rapid

accurate and non-destructive method for detecting Pb-contaminated mussels using near-infrared spectroscopy(NIRS)is proposed.The detection model of residualbased sparse extreme learning machine(RES-ELM)is constructed

and the influence of the maximum number of hidden neurons and sparsity parameter on the performance of the model is explored.The detection performance of RES-ELM

random pruning extreme learning machine(RP-ELM)and traditional extreme learning machine(ELM)is compared

and the changes in detection performance of these three models in the presence of Gaussian noise and class noise is analyzed.The results show that the proposed RES-ELM model exhibits better detection performance and robustness under different conditions.