Abstract: In the traditional sharing mode, the risk information of food full-process and all-information is sensitive and easy to leak, the self-interest and complete autonomy of centralized servers and malicious nodes easily give rise to malicious behaviors, cross-link interactions are difficult to deal with, and the single purposes causes low utilization of data. In view of these problems, this paper proposes a trusted sharing model for risk information of food full-process and all-information based on blockchain and federated learning. First, based on analysis of the information sharing needs of food companies and regulatory agencies, federal learning was integrated into a hierarchical blockchain architecture, and the relay concept was used to construct the trusted sharing model. Next, the federated learning process was divided into horizontal and vertical federated learning processes based on the characteristics and categories of risk information, and two federated learning aggregation algorithms were used to achieve the aggregation of data. Then, in response to the different needs of sharing risk information within and between enterprises, and between enterprises and regulatory agencies, homomorphic encryption algorithms and zero-knowledge proof were utilized for hierarchical encryption and targeted trusted sharing of risk information with different sensitivities. Finally, this model was validated by applying it to food safety risk assessment based on the food risk information disclosure dataset and open-source platforms. The results showed that the trusted sharing model could meet the different risk information sharing needs of enterprises and regulatory agencies, and achieve the full utilization as well as safe, credible, efficient and accurate sharing of risk information, thereby strengthening the determination of food enterprises and regulatory agencies to share data, and promoting the development of trusted data sharing and food safety digitalization in the food industry.

Key words: food full-process and all-information; blockchain; federated learning; homomorphic encryption; trusted sharing