Accurate and efficient identification of crop diseases can enable farmers to take effective and targeted preventive measures in a timely manner, which is helpful to reduce the risk of yield reductions and economic losses caused by crop diseases. However, the recognition model that can achieve the effect of SOTA in other fields, especially in the application of rice disease identification, faces the challenge of insufficient available rice disease data, a limited range of disease varieties and low data quality. In this paper, a variety of classical convolutional neural networks are trained on two different datasets using transfer learning methods. We demonstrated that in addition to the optimization achieved through model structure, the training data set itself has an important impact on the training results. However, the scarcity of open-source agricultural data, coupled with the absence of a comprehensive open-source data sharing platform, remains a substantial obstacle. This issue is closely related to the difficulty and high cost of obtaining high-quality agricultural data, low level of education of most employees, underdeveloped distributed training systems and unsecured data security. To solve those challenges, this paper proposed a novel idea to construct an agricultural data sharing platform based on federated learning framework, aiming to address the deficiency of high-quality data in agricultural field training.