A feasibility study of drone based identification and spot herbicide treatment of blackgrass in cereal fields

The experiment comprises over 40 ha of experimental plots with different blackgrass infestation levels that can be utilised for testing the described methods. Farmers and advisors have a central role in the project from day one. Together with the research team they will evaluate how to fit the developed methods fit into current farming processes as well as to evaluate the economic and ecological sustainability.

Blackgrass is one of the major weeds in much of Europe and has spread in the main Swedish winter cereal growing areas over the last two decades. In southern Sweden, the infestation level now seriously threatens the cultivation of winter annual crops. The problem is further exacerbated by the rapid selection of herbicide-resistant populations, which limits chemical control options. Recognizing blackgrass already at early infestation stages is important for taking countermeasures in time and maintaining the production in the long-term. Traditional weed recognition methods, such as field inspections, are time and resource consuming and generally cover only small parts of a field. As a result, farmers and advisors recognise blackgrass populations when large parts of a field are already infested. Determining geographic coordinates of blackgrass populations already at very early infestation stages is needed for targeted countermeasures that ideally focus only on the infested parts of a field instead of an economically and ecologically unsustainable area wide treatment. Drone-based remote sensing could be an efficient tool for detection of small blackgrass patches in large fields that have e.g. survived herbicide spraying, or which are at early infestation stages. We will utilize commercially available and affordable drones, equipped with a multispectral camera, which could capture images covering several visible and near infrared bands. Three methods will be further developed and tested for the detection of blackgrass and the differentiation from cereals. The first method involves training a discrimination model by extracting features from visual textures using high-resolution RGB true-colour (contains red, green and blue colours) images. Due to the high similarities between blackgrass and cereals, differentiation solely based on texture differences might be challenging. Taking advantage of the circumstance that cereals are typically grown in rows with a specific distance between them, the second method will be identifying crop rows in drone images and recognizing plants outside the rows as weeds. In the third method, we will calculate vegetation indices from drone multispectral images in order to assess their suitability for distinguishing blackgrass from cereals. The best-performed method will be applied to produce management maps that can be used for the targeted and georeferenced use of chemical and non-chemical tools for blackgrass control (spot treatment). Different drone-flying altitudes and image acquisition time across growing season will be applied to test the most suitable height and time for weed detection. In the second part of this project, we are going to test a drone-based herbicide application system, utilising the previously generated management maps. Most of the market available pesticide sprayers are not capable of spot spraying. Spraying by drone is time and energy efficient, causes no disturbance to the crop and significantly reduces the sprayed area to where the actual weed patches are located in the field. However, due to the legislative restriction, spraying chemicals from air is prohibited in Sweden. Upon necessary permissions, we will adopt a commercially available herbicide-spraying drone to test spot spraying. Economic and environmental analyses will be implemented to compare this method to traditional tractor pulled spraying systems. We will utilise an already established large-scale field experiment in Skåne where sustainable management strategies for blackgrass are currently tested.

Projektägare: Sveriges Lantbruksuniversitet
Projektslut: 2027-12-31
Budget: 2,1 miljoner kr
Kontakt: Junxiang Peng, 076-1358793