This paper aims to explore the response of Fourier transform mid-infrared (FT-MIR) spectra to the changes of the main components in wheat scab with infected different grades and to realize a non-destructive detection of grades of wheat scab based on FT-MIR spectroscopy combined with Sparse Representation based Classification algorithms. The FT-MIR spectra of 95 wheat samples infected with different grades of wheat scab samples were collected in 4 000~400 cm-1. The sensitive wavelengths in the FT-MIR spectra of wheat samples were selected by X-loading Weights and Random Forest algorithms, and Sparse Representation based Classification algorithms were used to build models to predict grades of wheat scab. The results showed that the characteristic wavelengths selected by XLW algorithm and RF algorithm achieved an accuracy of more than 90% for each qualitative analysis model, thus, the characteristic wavelength extraction algorithms could effectively simplify the model and improve efficiency. RF-SRC model had the best results, because the accuracy of the modeling set was 97% and the accuracy of the test data set was 96%. Being infected different grade wheat scab could cause the change of the content of water, starch, cellulose, soluble nitrogen , protein and fat in wheat samples. The characteristic wavelength selected by the RF algorithm could reflect the difference of the spectral characteristics of the FT-MIR spectra of these materials, so the grades discrimination of wheat scab by the RF-SRC model can achieve the best effect. Therefore, it is feasible to distinguish the grades of FHB in Wheat by using FT-MIR spectroscopy and pattern recognition method. This paper explained the mechanism of measuring the grades of FHB in Wheat by FT-MIR.