In order to achieve real-time variable dressing in wheat jointing stage, spectral data processing strategy and expected fertilizer quantity calculation model were studied. The variable rate fertilization expert decision system was developed based on crop canopy real-time spectral information (Normalized differential vegetation index, NDVI). Combined with NDVI information, actual traveling speed of dressing machine and feedback amount of fertilizer, the dual-channel independently controlled the speed and the opening of institutions to real-time adjust the amount of fertilization. The comparative experiments of ratio fertilization and traditional fertilization for winter wheat with different base fertilizer rates in Jiangsu Huanghai Farm were carried out. In order to improve the efficiency of expert decision system, the optimal data processing strategy was analyzed. Results showed that the best data processing frequency of the expert decision system was 1Hz. Using more fertilizers was beneficial to promote the winter wheat individual growth, but it was not conducive to the structure of the group. The variation coefficient of NDVI in control area was the largest, as high as 11.32%. Variable fertilizer could effectively reduce the variation cofficient of NDVI, decreased by up to 3.57% and the average of 1.46%. The fertilization precision can reach above 90%. Variable rate fertilizer applied more nitrogen (28kg/hm2) than quantitative fertilization. Variable rate fertilization would help to improve wheat population structure and reduce the yield differences. Therefore, the expert decision system was favorable, and could achieve precise fertilization in the precision agriculture.