Analysis of metadata standards for the exchange of image datasets and algorithms in the agricultural domain : A metadata-oriented approach to identify minimum interoperability mechanisms for image data and deep learning algorithms that is used for vision-based applications in agriculture. Sprint Robotics Project PL4.0 WP7

Abstract

This report discusses the importance of precision agriculture in achieving sustainability goals and the need for a basis that considers different perspectives of a data space such as interoperability, scalability, security, transparency, and data ownership. The Towards Precision Agriculture 4.0 project aims to address these perspectives to provide better-informed management decisions for farmers and the ecosystem. The current study focuses on determining minimum interoperability mechanisms concerning the standardization of image data and deep learning algorithms for vision-based applications in weed management by robots. The study adopts a metadata-oriented approach to make data and algorithms semantically interoperable and reuses existing knowledge from the Reference Model Agro (rmAgro). The results indicate the need for a balance between established standardization and agile standardization for supporting semantic interoperability, and the interoperability of preferred standards like Robot Operating System (ROS) and Open Neural Network Exchange (ONNX) is insufficient. The study results are useful for professionals and academia who work in the design and development of software for the farming business.