Abstract: In agricultural production, fertilization is one of the important means to improve crop yield and quality.However, excessive fertilization can cause various environmental problems, such as land and water pollution, affecting people’s health and the balance of the ecosystem.Therefore, it is of great significance to study the operating accuracy of variable type fertilizer applicator.The precision of fertilization is a very important link in agricultural production, which is directly related to the yield and quality of crops.Studies have shown that rational fertilization can improve nutrient utilization of crops, reduce fertilizer waste, and avoid the negative impact of excessive fertilizer use on the environment.In this study, fertilizer particles are taken as the research object, deep learning-based target detection is taken as the technical means, Python and NumPy library are used to realize HOG feature extraction, and visualization technology is combined with deep learning, aiming at target detection of fertilizer particles under field conditions.