Revolutionising Agriculture and Ensuring Food Safety with STELAR: Unleashing the Power of Agrifood Data Analysis
The new EU project STELAR (Spatio-TEmporal Linked data tools for the AgRi-food data space) is bringing together 9 partners from 5 European countries to redesign agrifood data analysis. To be more exact, the aim is to design, develop, evaluate, and showcase an innovative Knowledge Lake Management System (KLMS). This platform will enable simple and intelligent data discovery, AI-ready data, and semantic interoperability in smart agriculture and food safety applications.
Key Objectives of STELAR’s Activities
The STELAR KLMS aims to (semi-)automatically turn raw data lakes into knowledge lakes. It will support and enable easier access to FAIR (Findable, Accessible, Interoperable, Reusable) and AI-ready (high-quality, correctly labelled) data.
So what are STELAR’s objectives? And why did the need for a knowledge lake even emerge?
Facilitate and Improve Data Discovery and Reuse Through Automation
The need for better data discovery and reuse in the agrifood data space emerged because data spaces often hold large volumes of data of subpar quality and different formats. Additionally, metadata is often missing or incomplete. All this makes it challenging to find the right data, as well as a tedious manual process.
STELAR will help users faster find the data they need by automatically enhancing data descriptions.
This will be done through:
- Additional and high-quality metadata extracted from data catalogue content
- Different quality indicators (domain-specific ones included) taken from datasets
- Data summaries that make agrifood data analytics more energy-efficient over large data volumes
Support Data Linking and Interoperability
Data concerning the same entity is often scattered across different sources, resulting in name variations and spelling errors. With multiple data models, ontologies, and vocabularies coexisting, interoperability and data linking are much needed.
STELAR will use linked-data technologies to:
- Lower the manual labour needed through automated workflows
- Reach energy efficiency and scalability
- Ensure validity and efficiency of proposed methods
Better Automation of Data Annotation and Synthetic Data Generation
For robust Machine Learning, large amounts of labelled instances are needed. AI systems benefit from semantically annotated data to make informed decisions. But data labelling and annotation require both time and expertise of domain professionals. Many applications cannot afford that much dataset training. Also, data bias can have a negative impact on an AI application.
The project’s goal is to provide more AI-ready data. That is done by levelling up automation of data annotation and labelling, more specifically by:
- Employing supervised and active learning
- Adding techniques for bias detection and mitigation, as well as an explanation
- Lowering the lack of data and labelling by offering black-box and open-box synthetic data generators
Validate, Evaluate, and Demonstrate the Developed Tools in Real-World Use Cases
The food supply chain in the real world involves many factors and players. Factors such as Big Data, AI learning, and Machine Learning are transforming the field. Stakeholders such as producers, machinery manufacturers, governments, certification institutions, and others all have to exchange and share data. But this data is often coming from different places and in different forms.
The project will showcase how the KLMS helps overcome data management obstacles in the agrifood data analysis space. This will be demonstrated through three use cases, all three targeting different user needs:
Risk Prevention in Food Supply Lines
To protect consumers and ensure food safety, timely and precise risk assessments need to be done. And they are much easier to perform with interconnected data silos. At the moment, regulation data, farm management data, price data, weather data, and other kinds of information are stored in different types and dispersed all over the place. The use case will demonstrate how the KLMS helps generate AI-ready datasets for risk assessment.
Early Crop Growth Predictions
With the weather becoming increasingly hard to predict with traditional methods, this use case seeks to make predictions of food supply and warnings of food shortages through a more AI-informed process. The systems in place at the moment provide warnings that come too late in the season for farmers. STELAR’s KLMS will aim to combine data from different sources and incorporate it into the deep learning system and use it for physical modelling. That will help provide early crop growth predictions, while the crops are in their vegetative phase.
Timely Precision Farming Interventions
Using Earth Observation data along with farming technologies can help end-users with crop planning and land management. For instance, the STELAR KLMS seeks to combine meteorological data, IoT (Internet of Things) data from sensors and food processors, and best practices in order to come to conclusions about a crop product and help farmers develop agricultural strategies. For instance, the KLMS can provide data that informs food processing equipment on how to manage water accordingly.
Maximise Impact
Who Will Benefit from the STELAR KLMS?
Stakeholders from different parts of the agrifood data space can benefit from the Knowledge Lake Management System:
- Data-intensive SMEs (small and medium-sized enterprises) and SMEs moving to data-driven methods.
- Food processors, packagers, and distributors with the need to access food safety data
