Powder density, as a very important physical parameter, has a huge influence on the fluidity of powder, therefore, studying the characteristic of powder density is of great significance to powder processing, packaging, transportation, storage and so on. The hyperspectral scattering technique combined with moment method in hyperspectral scattering image feature extraction is used for non-destructive detection of bulk density of wheat flour. Hyperspectral scattering images of 474 wheat flour samples are acquired at the wavelength of 500-1000 nm. Images are pretreated to eliminate the noises, and the zeroth-order moment (ZOM) and the first-order moment (FOM) of scatter images are extracted. Finally, the ZOM, FOM, and their combination (Z-FOM) are used for developing bulk density prediction models using partial least squares (PLS) algorithm. The results demonstrate that the prediction model developed by Z-FOM achieved the optimal performance compared with ZOM or FOM. The PLS model using Z-FOM obtained 0.968 of the predicted correlation coefficient RP, and 3.95 of the residual prediction deviation. Experiments show that the moment method is an effective method for extracting features of the hyperspectral scattering images and can be used for high-accuracy detection of bulk density of wheat flour.