Abstract: The extra-root topdressing of rice at the tillering stage is one of the key steps in the management of rice production; it is also an important stage of fertiliser demand during the entire cycle of rice growth. The efficiency of extra-root topdressing directly affects the number of rice tillers and their growth in the middle and final stages. Due to the rapid advancement of the UAV technology in recent years, agricultural UAV are used for fertiliser spraying in the fields, which not only increased the rice yield but also reduced labour intensity and labour costs to a large extent, and greatly improved the efficiency of rice field management. In order to explore the use of UAV remote sensing to construct prescription maps to guide agricultural UAV to accurately topdressing rice at the tillering stage, realieze the field-scale nutritional diagnosis and UAV precise spraying, optimize fertilizer consumption, and ensure maximum rice yield,in this research, combining UAV remote sensing diagnosis with precision operation of agricultural UAV, UAV hyperspectral technology was used to establish the prescription maps of fertilization amount in rice tillering stage, combined with the operation parameters of agricultural UAV, grid division of fertilizing plots was carried out to determine the amount of precise fertilization, and precision fertilization was carried out by agricultural UAV. The consistent and desired end-member hyperspectral information of the ground features in the rice field were extracted to retrieve the nitrogen content of riceand a rice tillering stage fertilisation prescription map was established based in this, and the fertilization formula map of rice at tillering stage was established. According to the fertilizer amount prescription map, the operation parameters of agricultural UAV were set, and the plots to be fertilized were divided into grids to determine the spraying amount of each grid topdressing operation, and the precision topdressing was carried out by agricultural UAV. Dajiang spirit 4 RTK UAV was used to obtain the orthophoto image of the test fields with spatial information, the actual position of each topdressing grid was determined, and the variable spraying was realized by controlling the working voltage of the liquid medicine pump by PID algorithm. During the spraying process, droplet test cards were arranged on the ground at the same time to calculate the droplet coverage and other parameters such as droplet coverage rate. The results showed that five hyperspectral characteristic variables of rice were extracted in the 450-950 nm band by the method of feature band selection and feature extraction, the effects of rice nitrogen content inversion model constructed by Particle Swarm Optimization Extreme Learning Machine (PSO-ELM) was better than that of Extreme Learning Machine (ELM), and the coefficient of determination was 0.838 and the root mean square error was 0.466. The rice yield of UAV variable topdressing was basically the same as that of traditional topdressing, but the amount of pure nitrogen decreased by 27.34%.The study results can provide data and model basis for the precision variable topdressing of agricultural UAV in the tillering stage of rice in cold regions.