Objective To achieve high-precision detection of beef mince adulterated with pea protein, duck mince, chicken mince, and pork mince by combining hyperspectral technology with sparse hyperspectral feature selection.Methods The original spectral data of the beef mince samples is extracted and processed using standard normal variable transformation (SNV), multiplicative scatter correction (MSC), first-order differential (D1), and moving average (MA) preprocessing methods. A hyperspectral feature selection algorithm is designed based on sparse representation. This algorithm constructs a sparse dimensionality reduction framework and uses swarm intelligence optimization to optimize and solve the objective function of spectral feature selection. The spectral data dimensionality is reduced as much as possible while data diversity is maintained. Extreme learning machine classification (ELMC), random forest (RF), and support vector classification (SVC) adulteration detection models are built based on sparse hyperspectral feature selection are established, respectively. The effect of hyperspectral data combinations on the detection results is analyzed.Results Compared with the full wavelength, the classification accuracies of the three detection models based on sparse feature selection are increased by 2.33%, 1.86%, and 2.01%, respectively, superior to the ones established based on successive projections algorithm (SPA) feature extraction and competitive adaptive reweighted sampling (CARS) feature extraction. The combined spectral data processed by SNV and MSC has the highest detection and classification accuracy. Compared with that of the single spectral data, the classification accuracy is increased by 0.79%, 0.64%, and 0.65%, respectively.Conclusion The proposed method achieves effective detection of beef mince adulteration.