Success Story: Machine learning can help with weed detection
NCC presenting the success story
The Czech National Competence Centre (NCC) for High-Performance Computing (HPC) and Data Analysis (HPDA) is represented by IT4Innovations National Supercomputing Center at VSB – Technical University of Ostrava. Its mission is to analyse, implement, and coordinate all necessary activities and offer end users its services to meet their needs: from access to supercomputers and technology consulting to providing training for industry, public administration, and academia.
Industrial Organisations Involved:
The Ullmanna company was founded in 2019 by IT enthusiast Martin Ullmann and his father-in-law; the farmer Jindřich Ullmann. The mission of the company is to assist global farming with the necessary transition towards more sustainable and organic agricultural practices. The Ullmanna company focuses on decreasing or complete abandonment of pesticide utilization in modern farming, which is one of the key elements in addressing the urgent challenge of climate change and environmental degradation.
The vision of the Ullmanna company is to boost the efficiency and eco-friendliness of farming in conventional, organic, and sustainable farms. Nevertheless, it comes with the complex challenge of growing more organic crops without an enormous increase in the required workforce for weed control. Therefore, the Ullmanna company is developing an agricultural weeding machine that will enable in-line weed control by recognising the target crop using machine learning. This will lead to farming activities without the use of chemical sprays.
The researchers from the Advanced Data Analysis and Simulation Lab led by Jan Martinovič have participated in a unique project with the Ullmanna company. Both neural network architectures were optimised to run on the embedded device at 60 frames per second, allowing them to be used in real time on the weeding machine.
The Barbora supercomputer (using the TensorFlow framework), which is located at IT4Innovations, was used to train the object detection neural network based on datasets provided by Ullmann. This trained neural network was then tested in the field with a weeding machine prototype, and its performance was evaluated.
The cooperation on this project started thanks to the Digital Innovation Hub Ostrava, which connects the activities of IT4Innovations and the Moravian-Silesian Innovation Centre Ostrava, enabling SMEs to examine and solve their needs in the field of digitalisation, and which was financially supported by the Moravian-Silesian Region. The result of this cooperation led to further activities, e.g. obtaining a grant in the open call of the agROBOfood project funded by the H2020 programme, which will help to develop the commercial potential of the product.
To automatically identify crops, machine learning was employed, and a neural network was designed and trained on HPC infrastructure. Crop recognition, in this case sugar beet from weeds, allows the weeding machine to remove the weeds while not damaging the grown crop. The goal was to create a neural network model that can be used for inference on the weeding machine while meeting the constraining conditions of deployment (limited HW, inference speed, recognition accuracy). Two neural network architectures were designed and tested. The first was based on regression; it was easier to train and offered better performance, but only allowed the detection of one crop per photo. Subsequently, an architecture using object detection was proposed since it allows the detection of any number of objects in a photo.
- food production: 3x more profit per hectare;
- use of chemicals: 40% pesticide reduction;
- environment and health: no contamination of groundwater and no negative effect on workers´ health and safety;
- workforce: reduced time and required workforce.
SUCCESS STORY # HIGHLIGHTS:
- Keywords: eco-friendly farming, weeding machine, machine learning, object detection, neural network
- Industry sector: agriculture, weeding machine,
- Technology: machine learning, object detection, neural network
Jan Martinovic (email@example.com)
This project has received funding from the European High-Performance Computing Joint Undertaking (JU) under grant agreement No 951732. The JU receives support from the European Union’s Horizon 2020 research and innovation program and Germany, Bulgaria, Austria, Croatia, Cyprus, the Czech Republic, Denmark, Estonia, Finland, Greece, Hungary, Ireland, Italy, Lithuania, Latvia, Poland, Portugal, Romania, Slovenia, Spain, Sweden, the United Kingdom, France, the Netherlands, Belgium, Luxembourg, Slovakia, Norway, Switzerland, Turkey, Republic of North Macedonia, Iceland, Montenegro