Catalogue of Services

The Benchmarking & Testing Suite for Edge Hardware Systems delivers a set of tests designed to evaluate the performance, reliability, and functionality of edge hardware and its components under various operational conditions.

The tests can be compared to specific industry standards or the performance of other solutions available on the market. Examples of tests offered as part of the service:

- Environmental tests: Assessing devices' resistance to extreme conditions, such as temperature, humidity, and vibrations.

- Signal tests: Evaluating devices using GNSS signal generators, testing their resilience to interference or false signals.

- Network tests: Evaluate device performance within a prototype 5G network infrastructure.

- Functional tests: Assessing the capability of devices, such as remote PTZ (pan, tilt, zoom) cameras, to perform operational tasks in field conditions.

- Integration tests: Examining the cooperation of edge devices with sensors, AI systems, and their responses to data input failures. - Accuracy tests: Measuring the precision of sensors and control systems. Test results can be compared to specified standards or the performance of competitive solutions, enabling customers to better understand their devices' capabilities.

Desk assessment

Performance evaluation

Test design

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

Test setup

Curation and provision of annotated datasets

The service provides access to annotated data collected from devices tested as part of other project services. It includes both data annotation services (manual and automatic) and the provision of annotated databases or data packages. In addition, we offer disc space for temporary archiving of the collected data along with its annotation in cases where this is not possible at the client's site. As a result of the service, procedures for both manual and automatic annotation will be identified that are appropriate to the type of measurement data and the way it is collected and assembled. The methods of labelling and assigning specific data classes collected by the agriFoodTEF technical infrastructure will be proposed. Possibilities for documenting and storing data in datasets will be indicated. In addition, procedures for extracting selected features from stored research data can be offered.

Data harmonisation

Support for data harmonisation and integration, i.e., processing and transforming data from different sources into a standard common format. This process requires the specification and implementation of mappings between a data source model and a target (standard-based) model, the integration with other standard-based data sources (establishing links between data elements), and the provision of interfaces to access the harmonised data. The harmonisation can be performed as a one-time activity, e.g., with (semi-) static data sources, but more importantly in a continuous manner, whenever new data is becoming available. For the latter, a pipeline is defined with the specified mapping once, and then it can be re-executed (via CLI or API) as frequently as needed just by changing the input data.

Conformity assessment

Data augmentation

Wageningen University WUR

Evaluation for Ethical, Legal, and Social Aspects (ELSA Scan) for responsible AI development

What are the ethical, legal, and social aspects (ELSA) of AI-driven technology in agri-food? Mitigation of potential risks and finding opportunities are easier by identifying ELSA aspects in early development stages, when the technology is still in the making. Examples of ELSA aspects are autonomy, transparency, and bias, but also data privacy and other legal aspects. How does your AI technology deal with ELSA aspects? A combined survey and interview help AI developers to identify opportunities and issues with ELSA in the agri-food context from the perspective of end users and more broadly, for society. For example, to better align with the sustainability objectives of the AI-driven technology. The outcomes of the ELSA scan are a list of identified key ELSA aspects and high-level recommendations to further improve the AI technology.

ELSA assessment

Experimentation with synthetic data generation and data augmentation techniques

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

This service involves expanding training datasets using synthetic data and data augmentation techniques. This will use algorithms capable of generating synthetic data simulating certain aspects of existing datasets. Such an approach will lead to the enrichment of the existing training material. The service includes the creation of synthetic replicas based on real data, the combining of real and synthetic data, and also, for example, the creation of digital terrain maps based on ground data, which can be further enriched with synthetically generated details. As a result, we offer a comprehensive solution for more reliable development of datasets.

Data analysis

Data augmentation

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

Generation and provision of labelled datasets for benchmarking purposes

The service provides the ability to deliver to the client model training data sets with labels and metadata, which will later be used to conduct comparative analyses of AI algorithms by the target groups using the technologies under study. The datasets delivered to the client will be configured or customised in advance, as appropriate, according to the requirements and needs of the client of the service. Customisation processes will include configuring and adjusting the data and labelling according to the scopes of field operations, technological processes or communications to which the transferred data assets relate. These data sets include, but are not limited to, satellite and spatial data for remote sensing, phenological data for different crops, or synthetic data to expand training opportunities.

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

Providing demonstration testing on the test field of robotic or AI-driven solutions

Wielkopolska Agricultural Advisory Center in Poznań (WODR)

As part of the service, a testbed will be provided for product demonstration of robotic machines along with AI systems. The polygon will be configured and prepared to meet the customer's individual needs in terms of field operations, technological processes, and communication. The service provider will provide polygons with a variety of surfaces according to the customer's specifications, e.g., meadow, ploughed field. The service will include soil moisture control, continuous monitoring of the field, and all activities carried out on it. The service providers will provide the necessary assistance and infrastructure needed to support the fieldwork carried out by the surveyed sites. The service may also involve demonstration of AI systems used in apiaries.

Data analysis

Desk assessment

People training

Performance evaluation

Test design

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

Test setup

Provision of cloud-based data storage

Wageningen University WUR

Within the service we provide: - remote access to the disc space, allowing for safe storage of data (e.g., from sensors, photos from specialised cameras, weather data, satellite data, archive historical crop yields, etc.). - Depending on the needs, the available disc space can range from several gigabytes to tens, if not hundreds, of terabytes. - The stored data may be used directly by the customer or may be used in other services (this refers to using data for AI/ML model training, testing and validation for applications such as precision farming, yield prediction and pest detection). - The ability to store both structured (e.g., CSV, SQL) and unstructured data (e.g., images, videos)

Data analysis

Scanning as a Service for synthetic data generation and modelling

We offer a comprehensive 3D scanning and model preparation service tailored for agriculture and food applications, built on expertise developed through scanning diverse objects such as rapeseed plants, tomato seedlings, and fish. This service transforms real-world items into accurate, high-resolution 3D models, ready for use in synthetic data generation, robotics simulations, and plant phenotyping. Using tested scanning approaches—including photogrammetry with stationary or rotating cameras, DSLR, and smartphone captures—we can reliably acquire detailed image sets. Using these 3D models, your AI detection algorithms can be tested on high-quality 3D data. The data can be used to test detection algorithms, or it can be used in complete virtual environments to test robotic applications. Beyond static models, dynamic functionality can be added by assigning physics properties and enabling randomisation—altering geometry, texture, and patterns to expand dataset diversity. This allows models to be seamlessly integrated into simulation environments for testing AI, testing robotic systems, or conducting virtual experiments to test and improve the systems. Whether for research, product development, or automation, our solution delivers a scalable, future-ready platform for realistic and adaptable 3D tests.

Lukasiewicz Poznanski Instytut Technologiczny (L-PIT)

Technology advancement and readiness assessment

Wielkopolska Agricultural Advisory Center in Poznań (WODR)

This service helps you understand where your AI-powered product, system, or software stands in terms of technological maturity. We conduct a basic, yet structured evaluation of your solution—including physical testing of selected components and general system behaviour—to identify its current development stage and key areas that need improvement. We also assess the quality of documentation, system architecture, and software implementation. For example, when evaluating an apple-picking robot, we assess the readiness and integration of its core subsystems: the gripper mechanism, robotic arm kinematics and control, and the vision system responsible for detecting ripe fruit. We analyse areas where current solutions can be improved or require some redesign and whether the robot as a whole can effectively perform “scene ”cleaning”—removing ripe apples while preserving unripe ones and the plant structure. Based on our findings, we prepare a clear roadmap and a development plan with recommendations for the next steps, which may include further testing or refinement through other agrifoodTEF services. As part of the process, we also confirm the current Technology Readiness Level (TRL) of each major component. The outcome helps you focus your R&D work and understand what is still required to move your solution closer to full deployment and commercial readiness.

Conformity assessment

Testing of integrated AI system on a High-Performance Computing enviroment

Customers gain access to a remote environment with detailed manuals and a demo account. Additionally, high-speed storage and networking are available for the customer, allowing for fast data exchange. The environment offers usage of a containerisation approach, SSH access and user support. Thanks to this service, customer experiments will be supported with the use of high-end testing infrastructure (reinforced with the Nvidia H100 cluster). This service is also AI solution agnostic – each application or architecture dedicated to agrifood technology could gain an advantage from access to a testing and training environment.

AI model training

Testing security and robustness of AI systems

Our service offers comprehensive testing for the security and robustness of AI products, tailored to meet the specific requirements of our customers across various operational fields, technological processes, and communication systems. We specialise in conducting thorough security audits, penetration tests, vulnerability scans and code reviews to identify and mitigate potential security threats. By leveraging advanced testing techniques such as automated and manual pentesting, fuzz testing or hybrid source code reviews, we ensure that AI products are robust against cyber threats and/or adversarial attacks before they go to market. Our adaptable approach allows us to customise each service to align with the unique security needs of each customer, providing them with a high level of protection and confidence in their AI technologies.

Cybersecurity

Performance evaluation

Test design