Agriculture, as an essential part of food production, belongs to the Critical Infrastructures (CRITIS). Accordingly, the systems used must be designed for fail-safe operation. This also applies to the software used in agricultural operations, which must meet security and resilience criteria. However, there is an increase in software that requires a permanent Internet connection, i.e., a stable connection to servers or cloud applications is required for operation. This represents a significant vulnerability in terms of resilience and can lead to significant problems in the event of a telecommunications infrastructure failure.

With developments from the field of Resilient Smart Farming (RSF), we show how data storage can be designed according to the offline-first principle. A central building block here is Resilient Edge Computing (REC) and the developed HofBox: a mini-server that handles data management on the farm and implements it using innovative open source-based container technology (Open Horizon). This will make further use cases within the agricultural production and value chain realistic and feasible in the future through public-private partnership models. For the first time, we can manage and deploy containerized software not only on local computers and servers (Raspberry Pi) but also on mobile edge such as smartphones. This can have a positive impact on the performance of AI-based models, especially when internet availability is poor.

An innovative new feature is therefore management and deployment of containerized software on mobile devices. This significantly increases the flexibility for the use of containerized applications. Administration of the software used on servers, HofBoxes and mobile devices is made much easier. This makes it very flexible to use this software on different end devices across the entire hybrid cloud infrastructure. The scalability and flexibility within a hybrid cloud infrastructure is significantly increased through the use of AI tools. The ability to distribute the computing power individually according to the needs of the applications opens up a wide range of possible applications for AI in agriculture. We can show that complex AI-based models work in containerized software, e.g. for image recognition of weeds on mobile devices, and will significantly influence future developments.