To achieve the quick nondestructive detection of hazelnut protein, a near infrared spectroscopy and interval random frog algorithm-based hazelnut protein detection model is proposed in this paper. After extracting the near infrared spectral data of hazelnut, first-order derivative and standard normal variable transformation preprocessing on the hazelnut spectral data is performed. Considering the uncertainty of the initial subset of the random frog algorithm and the final band number threshold's uncertainty, an interval random frog algorithm is utilized to extract the characteristic band, competitive adaptive reweighted sampling algorithm, and successive projections algorithm, and the original random frog algorithm are compared. Furthermore, a partial least squares regression model is developed based on the extracted feature bands. The experimental findings depict that the interval random frog algorithm had the best performance and the developed model is more stable when compared with other algorithms. The regression coefficient and root mean square error of the interval random frog algorithm for the cross-validation set are 0.9082 and 0.0178, respectively, and the regression coefficient and root mean square error for the test set are 0.8999 and 0.0372, respectively.