Project title:
National-level weed detection using deep learning algorithm and Sentinel-2 data
Project ID: 4000148694
Fund providing support: ESA (Europian Space Agency)
Total project amount (amount tendered): EUR 118,888
Support rate: 100%
Beneficiary’s name: Ulyssys IT Development and Consulting Ltd.
Beneficiary’s address: Lövőház street 39., 1024 Budapest
Date of project implementation start: 09.07.2025
Project implementation closing date: 14.07.2026.
The current Common Agricultural Policy (CAP), effective until 2027, requires each Member State to develop, implement, and operate an Area Monitoring System (AMS) to monitor all subsidized agricultural areas using Copernicus Sentinel satellite data or other data of at least equivalent value. A fundamental eligibility criterion for agricultural parcels is the prevention of the spread of herbaceous and woody weeds. Agricultural areas affected by extensive and dense weed patches become ineligible for CAP funding. Furthermore, weeds are estimated to cause a 34% loss in global yields, negatively impact public health, decrease the stability and ecological service function of grassland ecosystems, and strongly affect animal husbandry on grasslands.
The objective of our project is the reliable detection of weed-infested grassland areas using Sentinel-2 data, which—with a five-day revisit time—is excellently suited for countrywide analytical tasks. However, Sentinel-2 satellite imagery has limitations in distinguishing weed patches, primarily due to its spatial resolution. Therefore, the main goal of the project is to examine whether a countrywide weed map can be generated and weed-covered grasslands can be reliably identified based on Sentinel-2 data. To achieve this, we must investigate and test an appropriate image classification method to produce reliable and accurate weed maps for end-users.
Consequently, our project focuses on weed detection and, in a later phase, on the development of a near-real-time weed monitoring system. The aim is to establish a workflow that integrates deep learning-based classification for weed identification with parcel-level evaluation based on these results.