Variable rate application is an effective way to realize low-pollution and high-efficiency use of pesticides. Real-time and accurate identification of disease information is a key prerequisite that affects the development of variable application technology. At present, the identification of diseases is mainly determined by the method of artificial field sampling, which not only is time-consuming and laborious, but also has disadvantages such as poor representativeness, strong subjectivity, and poor timeliness. Rice blast, which is the most severely affected by rice, is chosen as the object in this paper. Firstly, this paper proposes a rice blast online monitoring and real-time spraying system suitable for rice field sprayer. The pesticide will be sprayed according to the severity of the disease during the sprayer doing field inspections. Secondly, existing convolutional neural networks have slow convergence speeds in the identification of small samples of rice blast, which is prone to problems such as over-fitting. To solve the meaning problem, this paper proposes a rice blast monitoring algorithm based on migration learning. The knowledge learned by the VGG-16 network on the ImageNet image data set is transferred to this model, and a brand-new fully connected layer is designed. Finally, experiments show that the accuracy of the constructed rice blast identification model can reach 97.18%. It provides a reference for the intelligent diagnosis of diseases.