Conference Participation – XII Weeds Science Congress and symposium on herbicides and growth regulators

Date: 23-26 September 2024

Location: Belgrade, Serbia

Gerassimos G. Peteinatos

Computer vision and AI in weed detection: Current applications and future perspectives

Abstract

As the public’s concern for food security increases and stricter rules apply worldwide in order to restrict the use of herbicides in the agriculture-food chain, consumers are becoming less willing to accept chemical plant protection. A site-specific weed management technique involves applying a treatment to only the weed patches in order to achieve site-specific weed control. It is vital to identify plants and weeds when doing precision farming in order to spray herbicides on-site, robotically weed, and to control mechanical weeds with precision. There have been many different approaches to this problem over the last few years, but more work needs to be done to improve the speed, robustness, and accuracy of algorithms and recognition systems. Digital cameras and Artificial Neural Networks (ANNs) have rapidly developed in the past few years, providing new methods and tools also in agriculture and weed management. With computer vision and artificial intelligence (AI), weed detection methods have been significantly advanced. AI-powered cameras on tractors or drones can use image data to identify weeds and distinguish them from crops. Using deep learning algorithms trained on vast datasets of plant images, these systems are able to recognize different types of weed species at an early stage of their growth.

Autonomous robots, like Farming Revolution, Naïo Technologies, EcoRobotix, BlueRiver Technologies, etc, use cameras and AI algorithms to identify weeds and either mechanically remove them or apply targeted micro-doses of herbicide. These robots navigate fields independently, reducing labor and minimizing soil compaction. On the tractor implement side, there exist camera-guided hoeing systems that can distinguish between crops and weeds based on size, shape, and color. It is possible for these implements to adjust their tools in real time in order to get rid of weeds while leaving the crops untouched at the same time. Advanced systems combine detection with selective herbicide application, spraying only where weeds have been detected in order to provide spot spraying applications. Furthermore, interrow weed control as well as intrarow band spraying has also been developed and tested, reducing herbicide use as much as possible while maintaining effective weed control. There is continued development of both robotic and tractor-mounted detection systems, with ongoing research aimed at improving the accuracy of these systems across a variety of crop types and environments. There can be many technologies that will be implemented into agriculture in the coming years, including the development of better algorithms, higher recognition precision, faster and more robust implements, as well as fully autonomous systems or highly sophisticated systems.

Conference Participation – 22nd Scientific Conference of the Hellenic Weed Science Society

Date: 19–21 March 2025

Location: Arta, Greece

M. Tserioni, R. J. Van-Esdonk, D. Chachalis, K. Ferentinos, G. G. Petinatos

Utilizing Artificial Neural Networks (ANNs) for Accurate Weed Detection in Rice Cultivation

Abstract

Weeds pose a significant threat to rice cultivation, reducing yield and production quality. Specific species, such as red rice (Oryza sativa, ORYZ), barnyardgrass (Echinochloa spp., ECHCG), and smallflower umbrella sedge (Cyperus difformis, CYPDI), stand out for their competitive ability, making their accurate identification essential. In this study, 30 pots of each weed species were established, along with cultivated rice (Indica long sperm / CL111 variety), in the greenhouse of the Benaki Phytopathological Institute. High-resolution photographs were taken at various weed development stages, creating an extensive dataset. This dataset was used to develop digital weed recognition models through advanced image analysis and machine learning techniques. Two different machine learning (ML) models were used for weed differentiation and classification. In the Hue-Saturation-Value (HSV) color space, separation between cultivated rice and weeds was performed using Support Vector Machines (SVM). In the Red-Green-Blue (RGB) color space, an evaluation was made of the advanced convolutional neural network (CNN) YOLOv8, which is used both for object detection within the image and for classification. In both models, system performance measurement was based on established evaluation metrics: Precision, Recall, mean Average Precision at 50% (mAP50), and F1-score. These metrics were calculated both for each weed type individually and for all weeds collectively. The distinction between cultivated rice and red rice or barnyardgrass was satisfactory in both cases, with better results from YOLOv8; however, the same cannot be said for smallflower umbrella sedge. The use of these methods contributes significantly to improving the accuracy of weed detection systems, bridging the gap between laboratory conditions and field applications. The result is a more efficient and sustainable weed management approach in rice cultivation.

Conference Participation-20th European Weed Research Society Symposium

Date: 1-4 July 2025

Location: Lleida, Spain

Tserioni, Z. M., Psomiadis, E., Oikonomou, A., Kavadias, A., Avramidou, M., Petinatos, G. G., Kopaka, C., Petraki, A., Chachalis, D.

Advanced Approaches for Monitoring and Managing Weeds in Rice Cultivation.

Abstract

This study focuses on addressing the challenges posed by weeds in rice cultivation through the application of innovative methods that utilize ground-based and aerial multispectral sensors, as well as high-resolution satellite imagery. Particular emphasis is placed on estimating the density of three major weed species: Oryza sativa (ORYZ), Echinochloa spp. (ECHCG), and Cyperus difformis (CYPDI), which pose a significant threat to rice production due to their competitive ability. The research was conducted in the largest rice cultivation region in Greece (Chalastra, Thessaloniki) during the 2023–2024 growing season, in three experimental fields, each approximately 20 hectares in size. An advanced UAV (quadcopter drone) equipped with a MicaSense RedEdge-MX multispectral camera (5 bands: R, G, B, RedEdge, NIR) was used for data collection. UAV flights produced high-resolution orthophotomaps (3.5 cm/px), while weed density classification was performed in ArcMap, aiding in the spatial categorization of weed density in the fields. Specific vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Red Edge Index (NDRE), were calculated, enabling correlation of weed density with the data derived from the orthophotomaps. Satellite images were obtained from Sentinel-2A (12 bands with resolutions ranging from 10 m to 60 m), in chronological correspondence with UAV flights at two time points: August 1–3 (BBCH 57–59) and September 11–13 (BBCH 77). In the pilot area, 48 samples were collected using a 0.5 × 0.5 m sampling frame (September 11–13) to estimate weed biomass as a percentage of total crop biomass: low (5.22%), medium (18.3%), and high (31.1%). These data provided significant input for correlating digital weed detection models with the vegetation indices derived from UAV and satellite imagery. This study highlights the importance of integrating digital tools, such as UAV-based multispectral imaging and satellite data, for precise weed mapping and management. The use of these technologies supports sustainable rice cultivation, promotes the rational and timely use of herbicides, and contributes to limiting weed resistance to herbicides.

Conference Participation – FOOD2030 Networks Conference 2025

Date: 2–4 December 2025

Location: Copenhagen, Denmark

Demosthenis Chachalis, Aggeliki Petraki, Nikolina Vidali, Dimitris Vlotsos, Athina Motsenigou, Gerasimos Peteinatos, Alma Balestrazzi 

Participatory Pathways for Agroecological Transition in Mediterranean and European Food Systems

Abstract

Over the last five years, our research group in Benaki Phytopathological Institute (BPI) has been involved either as coordinator or a key partner in several Mediterranean projects (i.e. PRIMA-ZeroParasitic, PRIMA-Benefit-Med, and the Greek project ARTs-rice) aiming to advancing participatory frameworks for agroecological innovation in various cropping systems. These projects established multi-actor Living Labs and Communities of Practice bring together farmers, researchers, SMEs, and policymakers to co-design and test agroecological solutions in crop management, input reduction, and ecosystem restoration. Through iterative learning cycles, participatory field trials, and digital tools for knowledge exchange, the projects strengthened local innovation capacity and demonstrated the value of farmer-scientist collaboration for adaptive management under changing climatic and socio-economic conditions. Building upon this foundation, the newly launched Horizon Europe project PROSPER (2025–2029) expands this participatory legacy into a pan-European framework for resilient and biodiversity-friendly agroecosystems, by focusing on orphan legumes. PROSPER integrates the methodological and social innovations developed under previous initiatives into a cohesive platform for co-creation, behavioral adoption, and policy uptake. The project mobilizes Living Labs as engines of transition, linking agroecological science with digital decision support, circular practices, and market incentives. By reinforcing transnational collaboration and farmer engagement, PROSPER aims to accelerate the scaling of agroecological solutions, contributing directly to the European Green Deal, Farm to Fork, and Zero-Pesticide ambitions for sustainable food systems.