Validation of AI-based models of crop fields

A service providing experimental fields and data collection tools to test, validate, and calibrate predictive models, algorithms, and AI-driven solutions for precision agriculture under real-world conditions. 

Interested in this service? Contact us at roberto.garcia[at]udl.cat 

 

Overview

We provide experimentation fields for herbaceous crops (like wheat, barley, maize, soybean, sunflower, pea, rapeseed, camelina, and more), along with sample laboratory analysis or crop characterisation if required. 

All combined are used as testbeds for data acquisition throughout the crops’ growth cycle for training predictive models or helping the company’s development and improvement of their AI-based algorithms for agronomy applications, such as automatic weed detection for precision spraying solutions, biomass and yield estimation, phenological characterisation, etc. 

This service also helps the validation of agricultural technology solutions based on, but not limited to, systems that require physical testing, such as proximal remote sensing technologies, including UAVs (unmanned aerial vehicles, or drones), computer vision systems for weeding machines, spraying applicators, etc. 

Download factsheet description

More about the service

Discover more about our service, including how it can benefit you, the delivery process, and the options for customisation tailored to your specific needs!

This service is designed to meet the needs of companies to validate or refine their agricultural technologies under real-world conditions. It enables the adjustment and adaptation of predictive models, algorithms, or systems to specific environments and crops by providing high-quality field datasets and testing setups. Before using this service, you might have an algorithm or model that works in a controlled setting but lacks calibration for a specific real-field variability. After the service, you will have access to comprehensive datasets, including field imagery, crop characterisation, and environmental parameters, enabling you to fine-tune your solution for optimal performance. To illustrate, if your algorithm identifies phenological stages in soybeans but requires adaptation for maize identification, our service enables you to gather and link real-time drone imagery with crop-specific data. This guarantees that your solution is not only functional but also accurate and reliable under diverse conditions. 

Data collection will be performed throughout the crop cycle, for example, with drone flights or other imagery systems, and representative sampling will be conducted during key phenological stages. Sampling and laboratory analysis will provide relevant agronomic parameters, such as yield, biomass, nutritional composition (water, nitrogen, or carbon content), or any other physiological characterisation. The datasets will be delivered alongside detailed reports adapted to the customer’s needs. These outputs enable the customer to calibrate their algorithms or models effectively. The duration of the service depends on the specific crop and its growth cycle, typically spanning several months. 

This service can be partially customisable due to the availability of crops currently being grown in the experimental fields, which are determined by ongoing agronomic trials and seasonal cycles. While adjustments to crop species, varieties, or planting layouts can sometimes be accommodated, these must align with operational constraints and be planned well in advance. Customers should communicate their testing requirements early to ensure the service meets their objectives. Any limitations regarding the feasibility of customisations will be discussed during the planning phase. • Measured laboratory parameters can be customised. • Field adjustments may include crop species and varieties selection, planting configurations, or sampling strategies. 
Location
At user's premises
Spain
Type of Sector
Arable farming
Type of service
AI model training
Collection of test data
Data analysis
Performance evaluation
Provision of datasets
Test execution
Accepted type of products
Physical system
Software or AI model
Other