Based on a multispectral image angle fusion. Multispectral images were obtained from 10° to 90 °at 10° intervals. Multispectral image angle fusion and the region of interest (ROI) were extracted using ENVI5.1 software to obtain the multispectral data. The Pearson correlation analysis of the spectral reflectance, band, and relative azimuth found that both the band and relative azimuth were extremely significantly correlated with the spectral reflectance, and the relative azimuth correlation coefficient of 0.1 is greater than the band correlation coefficient of 0.053. Therefore, it is necessary to add the relative azimuth factors in the modeling process. Using standard normal variable transformation (SNV), mean centralization transformation (MC), convolution smoothing treatment (S_G), normalization treatment (Nor), partial least squares regression (PLSR) to evaluate the full band set correlation coefficient (Rc), prediction set correlation coefficient (Rp), correction set root mean square error (RMSEC) and prediction set root mean square error (RMSEP) to explore the effect of the model. The results show that the prediction effect of the established PLSR and SVM models is significantly improved after adopting the angle fusion treatment. The optimal model is a partial least squares regression model (AF-PLSR) with Rc of 0.936, RMSEC of 0.298, Rp of 0.901, RMSEP of 0.285; the optimal prediction model is the support vector machine model (AF-SVM), Rc is 0.894, 0.527, 0.376;Rp is 0.830, 0.901, and RMSEP is 0.532, 0.379 respectively. Angle fusion combines the spectral data from different angles together to obtain more abundant information than a single angle and a more perfect spectral information. The established detection model has a higher accuracy. The results proved that it is feasible to predict the water content, hardness, and sugar content of Korlas fragrant pear based on the multispectral image angle fusion technology. The results provide a new idea for improving MMS and HMS NDE accuracy.