Abstract: Iron grade detection is a key technology in iron ore powder production. In view of the shortcomings of current iron grade detection methods,a mathematical method of iron grade inversion by hyperspectral curve is constructed,which has the advantages of fast detection speed and no secondary pollution. In this paper,54 iron ore samples from Shirengou were selected to measure the iron grade by chemical method,and the hyperspectral curves of the samples were measured by ASD FieldSpec4 spectrometer. It is found that iron ions in iron ore powder exhibit strong absorption characteristics at 526,713,905, 998 and 1 100 nm,and the absorption depth of samples with different iron grades is obviously different,and there is obvious interference noise. Based on the strong correlation between hyperspectral absorption characteristics of samples and iron grade,the Least Squares Method,Robust Estimation and Ridge Estimation were used to model hyperspectral iron grade. The experimental results show that the accuracy of iron grade inversion is different under different constraint rules. The accuracy of the least square estimation inversion is low,the robust estimation can reduce the interference of a few anomalous spectral values,and the ridge estimation can effectively avoid the influence of complex collinearity between parameters on the accuracy. Under the joint constraints of robust estimation and ridge estimation,the model has strong generalization ability and high accuracy of iron grade inversion,which can effectively improve the accuracy of iron grade inversion prediction,and can be used as a basic model for hyperspectral detection of iron grade for iron ore powder.