Spatial heterodyne Raman spectroscopy is a hyperspectral detection technology that has emerged in recent years. It has the characteristics of non-contact, fast, simple, repeatable and no sample preparation required for Raman spectroscopy, but also the advantages of high resolution, high throughput and no moving can detect signals in the characteristic wavelength center range of the measured realize the direct measurement of weak Raman optical signals. Due to the weak signal to be measured, the machining accuracy of optical components, and the errors caused by device packaging and instrument installation, the interferogram received by the spatial heterodyne Raman spectrometer (SHRS) will have uneven light intensity distribution, interference fringe tilt or distortion. Therefore, the accuracy of the spectrum signal obtained by the ordinary spectrum recovery method is reduced or even hard to be identified. According to the error characteristics of the interferogram detected by SHRS, the two-dimensional Fourier transform is applied to the spectral restoration of the SHRS interferogram and a method of extracting the strongest line direction spectrum based on two-dimensional frequency domain spectrum resampling is proposed. The extraction process is to perform a two-dimensional Fourier transform on the collected interferogram of the target to obtain a two-dimensional spectrogram. By using the position information of the characteristic peaks of the two-dimensional spectrum signal of the single-wavelength or multi-wavelength light source collected by the same experimental system, the linear equation of the maximum direction of the light information intensity is obtained by fitting. According to the coordinate position of the intersection point of the straight line and each column of the target two-dimensional spectrogram, the pixels and the weight contributed by the resampling are determined. All columns of pixels is resampled along the fitted linear equation to obtain the final spectral signal. The method is applied to the clover interferogram data, and the recovered spectrum is compared with those obtained by other methods. The results show: compared with the one-dimensional row average spectroscopy method, the spectrum obtained by this method has a more obvious signal intensity in the center area of the detection eliminates the influence of the noise at the same time; compared with the direct extraction method of the two-dimensional spectrum center row, it slightly improves the recovered spectrum. However, due to the influence of the y component of the interference fringe, the final spectrum obtained by resampling along the strongest direction has a narrower half-width of the main peak and a smaller side frequency noise intensity. If the influence of the y component increases, the spectrum restoration effect of this method will become more obvious. This method is a useful supplement and an attempt for data processing of spatial heterodyne Raman spectroscopy.