Nitrogen topdressing is a vital agrotechnical measure to boost tillering process and improve the population structure of wheat stalks. However, uniform nitrogen topdressing is apt to cause excessive application and low agronomic efficiency of nitrogen fertilizer. However, variable-rate fertilization can solve the contradiction between individual development and formation of population structure of wheat stalks, by decreasing the application of nitrogen fertilizer according to the actual growing status of wheat stalks in the field. The key technology to improve the population structure of wheat stalks using variable-rate nitrogen topdressing is to accurately obtain the information of spatial variations of wheat stalk numbers at field scale. Thus, with the objectives of fertilizer reduction and yield increasing, this research studies wheat growth status in the tillering stage to invert and regulate population densities of wheat stalks at field scale. Firstly, a DJI mini 2 with a CMOS (Complementary Metal Oxide Semiconductors) image sensor is utilized to acquire visible-band imagery of wheat from the experimental field. Secondly, vegetation indices of ratio type such asVDVI (Visible-band Difference Vegetation Index), NGRDI(Normalized Green-Red Difference Index), NGBDI (Normalized Green-Blue Difference Index), and RGRI (Ratio Green-Red Index)were calculated out of the visible-band images, in order to highlight vegetation features and reduce the impact of uneven light intensity on remote sensing images. Furthermore, FVC (Fractional Vegetation Coverage), which indicates the growth vigor of both individuals of wheat tillers and stalk population as a whole, was calculated based on the VDVI map. Subsequently, a BP (Backward Propagation) Neural Network prediction model was built to quantitatively invert wheat stalk density, using FGV, VDVI, NGRDI, NGBDI, and RGRI the input layer, and ground truth samples of wheat stalk densities as output layer. Upon completion of the BP Neural Network training, weight and threshold values of the prediction model were obtained, and a validation experiment was implemented. The result of the validation experiment showed that the RMSE (Root Mean Square Error) and MAPE (Mean Absolute Percentage Error) of the BP Neural Network prediction model is 19 and 3.62%, respectively. Compared with the average value of 635 of the wheat stalk density’s ground truth values, the BP Neural Network model has extraordinary wheat stalk density prediction accuracy. The statistical data of the inversed wheat stalk density at field scale indicated that the area of wheat stalk density below 500 stalks·m^-2, between 501~800 stalks·m^-2, and above 800 stalks·m^-2 accounted for 6.67%, 74.67%, 18.66%, respectively, which provided data support for variable-rate nitrogen topdressing. The implementation of this research under the background of “negative growth of fertilizer usage” proposed by the state is the actual demand for developing resource-saving and environment-friendly Green Agriculture. The research results provide new approaches and technical support for digitalization of wheat plantation, theoretical basis and data support for creating a high and stable yield of wheat in a large area, which is of great scientific significance.