• Rural Digital Europe
• Publications close
• Ghent University Academic Bibliogra... close

Weed and crop segmentation is becoming an increasingly integral part of precision farming that leverages the current computer vision and deep learning technologies. Research has been extensively carried out based on images captured with a camera from various platforms. Unmanned aerial vehicles (UAVs) and ground-based vehicles including agricultural robots are the two popular platforms for data collection in fields. They all contribute to site-specific weed management (SSWM) to maintain crop yield. Currently, the data from these two platforms is processed separately, though sharing the same semantic objects (weed and crop). In our paper, we have proposed a novel method with a new deep learning-based model and the enhanced data augmentation pipeline to train field images alone and subsequently predict both field images and UAV images for weed segmentation and mapping. The network learning process is visualized by feature maps at shallow and deep layers. The results show that the mean intersection of union (IOU) values of the segmentation for the crop (maize), weeds, and soil background in the developed model for the field dataset are 0.744, 0.577, 0.979, respectively, and the performance of aerial images from an UAV with the same model, the IOU values of the segmentation for the crop (maize), weeds and soil background are 0.596, 0.407, and 0.875, respectively. To estimate the effect on the use of plant protection agents, we quantify the relationship between herbicide spraying saving rate and grid size (spraying resolution) based on the predicted weed map. The spraying saving rate is up to 90 % when the spraying resolution is at 1.78 x 1.78 cm2. The study shows that the developed deep convolutional neural network could be used to classify weeds from both field and aerial images and delivers satisfactory results. To achieve this performance, it is crucial to perform preprocessing techniques that reduce dataset differences between two distinct domains.

In the Internet of Things (IoT), semantic IoT platforms are often used to solve the challenges associated with the real-time integration of heterogeneous IoT sensor data, domain knowledge and context information. Existing platforms mostly have a static distribution and configuration of queries deployed on the platform's stream processing components. In contrast, the environmental context in which queries are deployed has a very dynamic nature: real-world set-ups involve varying tasks, device resource usage, networking conditions, etc. To solve this mismatch, this paper presents DIVIDE, an IoT platform component built on Semantic Web technologies. DIVIDE has a generic design containing multiple subcomponents that monitor the environment across a cascading architecture. By monitoring the use case context, DIVIDE adaptively derives the appropriate stream processing queries in a context-aware way. Using a Local Monitor deployed on edge devices, situational context parameters are measured and aggregated. The Meta Model allows modeling these measurements, and meta-information about devices and deployed stream processing queries. Through the definition of application-specific Global Monitor queries that are continuously evaluated centrally on the Meta Model, end users can dynamically configure how the situational context should influence the window parameter configuration and distribution of queries in the network. The paper evaluates a first implementation of DIVIDE on a homecare monitoring use case. The results show how DIVIDE can successfully adapt to varying device and networking conditions, taking into account the use case requirements. This way, DIVIDE allows better balancing use case specific trade-offs and achieves more efficient stream processing.

Irrigation plays a crucial role in maize cultivation, as watering is essential for optimizing crop yield and quality, particularly given maize’s sensitivity to soil moisture variations. In the current study, a hybrid Long Short-Term Memory (LSTM) approach is presented aiming to predict irrigation scheduling in maize fields in Bursa, Turkey. A critical aspect of the study was the use of the Aquacrop 7.0 model to simulate soil moisture content (MC) data due to data limitations in the investigated fields. This simulation model, developed by the Food and Agriculture Organization (FAO), helped overcome gaps in soil sensor data, enhancing the LSTM model’s predictions. The LSTM model was trained and tuned using a combination of soil, weather, and satellite-based plant vegetation data in order to predict soil moisture content (MC) reductions. The study’s results indicated that the LSTM model, supported by Aquacrop 7.0 simulations, was effective in predicting MC reduction across various time phases of the maize growing season, attaining R2 values ranging from 0.8163 to 0.9181 for Field 1 and from 0.7602 to 0.8417 for Field 2, demonstrating the potential of this approach for precise and efficient agricultural irrigation practices.

AbstractQuantitative structural models (QSMs) are frequently used to simplify single tree point clouds obtained by terrestrial laser scanning (TLS). QSMs use geometric primitives to derive topological and volumetric information about trees. Previous studies have shown a high agreement between TLS and QSM total volume estimates alongside field measured data for whole trees. Although already broadly applied, the uncertainties of the combination of TLS and QSM modelling are still largely unexplored. In our study, we investigated the effect of scanning distance on length and volume estimates of branches when deriving QSMs from TLS data. We scanned ten European beech (Fagus sylvatica L.) branches with an average length of 2.6 m. The branches were scanned from distances ranging from 5 to 45 m at step intervals of 5 m from three scan positions each. Twelve close-range scans were performed as a benchmark. For each distance and branch, QSMs were derived. We found that with increasing distance, the point cloud density and the cumulative length of the reconstructed branches decreased, whereas individual volumes increased. Dependent on the QSM hyperparameters, at a scanning distance of 45 m, cumulative branch length was on average underestimated by − 75%, while branch volume was overestimated by up to + 539%. We assume that the high deviations are related to point cloud quality. As the scanning distance increases, the size of the individual laser footprints and the distances between them increase, making it more difficult to fully capture small branches and to adjust suitable QSMs.

This paper sheds light on the integration of digitalisation in multifunctional and diversified agriculture. In this empirical work based on socio-cyber-physical assemblage thinking, first we found that on-site material and immaterial conditions like living on the farm, farm settings, and creative human work still matter and deserve more attention in responsible digital agriculture. Second, in our small sample we found that internet and cloud-computing are the main digital technologies used in on-farm diversification, particularly infrastructures-as-a-service and software-as-a-service. Third, when farmers introduce digital technologies in their diversification, three tentative configurations of socio-cyber-physical assemblages are outlined and can be further assessed in future research: punctiform, horizontal, and vertical configurations, or a combination thereof. For each configuration, we illustrate key features and examples. Nine clusters of factors were identified as affecting these assemblages. These factors could help researchers and practitioners to understand why the digitalisation of diversified farms can be so diverse and needs to be carefully adapted. For policymakers, these factors can also serve to qualify the setting up of conditions to steer integrated agri-food-rural policies towards on-farm diversification, including responsible digitalisation. The paper argues that digitalisation in agriculture can unfold through different strategies (minimising, mutualising, grounding, enlarging digitalisation) and requires holistic and ambitious policy commitments for sustainability. For example, the paper suggests the setting up of a new EU-wide target for increasing the share of farmers engaged in on-farm diversification through the Common Agricultural Policy or European Farm-to-Fork Strategy.

ispartof: PLANTS PEOPLE PLANET status: Published online

Terrestrial laser scanning (TLS) has been adopted as a feasible technique to characterize tree stems while the characterization of trees' branching architecture has remained less explored. In general, branching architecture refers to the spatial arrangement of branches and their characteristics that are important when exploring the eco-physiological functioning of trees or assessing tree biomass and wood quality. Our aim was to develop a point cloud processing method for identifying and segmenting individual branches from TLS point clouds. We applied a Cartesian-to-cylinder coordinate transformation and a simple morphological filtering for stem surface reconstruction and stem-branch separation. Then branch origins were identified as their intersections with the stem surface, and individual branches were segmented based on their connectivity with the branch origins. The method, implemented in MATLAB and openly available, was validated on a 0.4-ha mature and managed southern boreal forest stand. The branch identification performance was assessed based on visual interpretation of 364 randomly sampled stem sections from 100 Scots pine (Pinus sylvestris (L.)) trees that were inspected for branch identification accuracy. The results showed that the branches could only be identified up to the height where the stem could be reconstructed. For 90% of the trees, this threshold ranged between 59.3% and 81.2% relative tree heights. Branches located below this threshold were identified with a recall of 75%, a precision of 92%, and an F1-score of 0.82. Based on our study, it appears that in a managed Scots pine stand, most of the branches can be identified with the developed method for the most valuable stem part eligible for logwood. The findings obtained in this study promote the feasibility of using TLS in applications requiring detailed characterization of trees. The developed method can be further used in quantifying the characteristics of individual branches, which could be useful for biomass and wood quality assessment, for example.Graphical Abstract Peer reviewed