Conference Agenda

Botrytis cinerea causes major losses in tomato. We acquired a time-series imaging spectroscopy dataset of detached tomato leaves to quantify lesion development across ten genotypes (resistant and susceptible ). Leaves were imaged at 0, 24, 48, and 72 h post inoculation (12 leaves per genotype per time point; four droplets per leaf; VIS–NIR at all time points and the SWIR at 72 h.) Leaf and lesion regions were segmented with SAM3, using a stage-specific text-prompt strategy reflecting changes in lesion appearance over time. We define the halo as tissue surrounding a lesion that is visually asymptomatic at an earlier time point but later becomes part of the lesion area. To compare halo spectra despite leaf deformation, we use an inscribed-circle matching approach: the largest inscribed circle of the 72 h lesion is mapped onto earlier images to define a consistent halo region (future lesion area minus current lesion area). Calibrated reflectance spectra were extracted for necrotic lesion tissue, halo tissue, and surrounding asymptomatic tissue. Preliminary analyses indicate the halo is distinguishable from asymptomatic tissue at 48 h in VIS–NIR spectra, and lesion spectra show genotype-dependent differences in SWIR at 72 h.

Accurate image annotation is a fundamental yet time-consuming step in building datasets for training object detection models in precision agriculture. In this work, we compare manual and automatic annotation strategies on two datasets of vineyard images: one containing more than 2500 images of grape bunches and one containing more than 1500 images of inflorescences.

Manual annotations, consisting of bounding boxes, are compared against automatically generated ones obtained using SAM3, the latest iteration of the Segment Anything Model, employed in a zero-shot fashion without any task-specific fine-tuning. The comparison is conducted using standard evaluation metrics to assess the agreement between the two annotation strategies. Notably, while manual annotation is limited to bounding boxes, automatic methods like SAM3 produce high-quality instance segmentation masks, representing an added value at no significant additional cost. Our results suggest that zero-shot automatic annotation tools, such as SAM3 and similar foundation models, can serve as viable alternatives or complements to manual annotation pipelines, potentially reducing the human effort required while maintaining annotation quality. This work contributes to the broader discussion on scalable and efficient annotation workflows in precision agriculture applications.

The integration of robotic platforms equipped with ultraviolet-C (UVC) emitters is emerging as a promising strategy to reduce chemical inputs in horticultural systems. This study presents a preliminary evaluation of a robotic platform for UVC applications in open-field lettuce (Lactuca sativa L.), focusing on potential photophysiological side effects associated with different radiation doses.The experiment was conducted on five lettuce cultivars. Two strategies were compared: (i) single weekly UVC application and (ii) bi-weekly applications, both set at 200 J m⁻² per treatment. In addition to treatment frequency, the effect of increasing UVC dose was evaluated at 400, 600 and 800 J m⁻². The main objective was to assess the potential impact of UVC exposure on the photosynthetic apparatus.Photosynthetic performance was evaluated through chlorophyll fluorescence measurements using a fluorometer. Parameters related to Photosystem II efficiency were used to characterize short-term physiological responses under different treatment frequencies and radiation doses. Preliminary results indicate cultivar-dependent responses and dose-related effects on photosynthetic activity. These findings provide insights into the balance between operational efficiency and physiological safety when implementing robotic UVC treatments in lettuce cultivation. Future work will include a biological evaluation of pathogen suppression and crop performance to assess the agronomic potential of this technology.

Brussels sprout cultivation is hampered by the appearance of the Cabbage whitefly (Aleyrodes proletella). The larvae secrete honeydew, a feeding medium for Sooty mould, causing harvest losses of up to 50%. Control relies mainly on synthetic insecticides. Increasing resistance to, and limited availability of active ingredients necessitate sustainable alternatives.

This project evaluates whether whiteflies can be removed in conventional Brussels sprout systems using a suction machine. Suction alone proved insufficient because of the dense canopy, the sheltered position of whiteflies, and the risk of removing beneficial insects. Therefore, whiteflies must first be detached from leaves before suction.

Two field tests compared potential detachment methods. Applying an incoming air stream was most effective. On average, this approach removed 70% of the whitefly population, with efficacy varying between canopy layers.

Stakeholder requirements were collected and translated into a machine concept using the Reflective Interactive Design (RIO) approach. The resulting CAD machine design can serve as a new Integrated Pest Management (IPM) measure in Brussels sprout cultivation. The presentation will cover the final design, stakeholder analysis, field-test results, and a financial assessment of the machine in use. This work was conducted as part of an MSc thesis in Biosystems Engineering at Wageningen University.

Airborne fungal spores are key drivers of epidemic development in many high-value crops. In perennial and specialty systems such as hop, vineyards, or orchards, decision support often relies on weather-based models or manual spore assessments. Weather models estimate infection risk but do not directly reflect actual spore flight and may become less reliable under changing climatic conditions and pathogen phenology, while manual assessments are labor-intensive and spatially limited. Advances in automated aerosol monitoring and artificial intelligence (AI) enable new approaches for real-time spore detection and site-specific disease forecasting.

This contribution presents a technical classification of real-time aerosol traps and evaluates their applicability in agricultural early-warning systems. Volumetric samplers combined with automated digital microscopy are contrasted with optical in-flight measurement systems. Their performance is assessed using comparable criteria relevant for decision support, including detection reliability, temporal resolution, field robustness, and economic feasibility. Volumetric systems are compatible with established threshold-based services, whereas optical systems provide high-frequency data suited to research and model refinement.

Improved temporal and spatial resolution enhances the alignment between spore flight and infection conditions without requiring changes in application technology. Current limitations, validation gaps, and research needs are discussed in relation to agronomic and economic implementation.