ABSTRACT. The presentation will focus on the panorama of the IoT technology, the motivation and the requirements of sensory for precision agriculture will be highlighted. A state-of-the-art and the trend of multiscalar wireless sensor networks (WSN) will be presented. To illustrate the presentation some use cases developed by SMIR group of LIMOS laboratory including remote real-time demo will be shown. An assessment and open research issues such as robustness and bandwidth will be addressed to conclude the presentation.

ABSTRACT. Flavescence dorée (FD) is a quarantine disease with huge consequences on the vine-growing economy. Affected vines show foliar symptoms depending on the cultivar, including yellowing and leaf curving. In this study, we propose a Monte-Carlo framework aimed at testing Complete Space Randomness hypothesis of the FD repartition using a histogram similarity measure. The experimental dataset consists of 7 plots in the Bordeaux region, exhibiting strong differences in plot size, disease incidence and dead plants proportion. FD presence was checked on these plots during august 2016, in the form of gridded data and georeferenced shapefiles. A statistic based on the local distance between symptoms is computed for a set of random maps. The statistic is then computed on the test dataset and compared with the distribution of the statistic under the null hypothesis in order to compute the p-value associated with the CSR test. Results on real dataset show non-random repartition of the FD symptoms on some of the experiment plots. In the future, the proposed methodology will also be applied on datasets with other diseases, such as Grapevine Trunk Diseases (e.g. Esca and Black Dead Arm) whose spatio-temporal repartition is not yet well understood.

ABSTRACT. Predicting grain yield a few weeks before harvest is strategic for the cooperatives which collect and store grains. A range of methods have been developed to predict grain yield and quality (oil, protein) for various field crops at different spatial and temporal scales. Here, we will combine remote sensing from satellite and statistical models to predict sunflower yield during grain filling. In 2014 and 2015, 187 sunflower fields were surveyed in Midi-Pyrénées region (SW France) throughout the season. Green area index (GAI) was estimated by the inversion of radiative transfer model with BVnet from satellite images. Two variables were calculated from GAI with time: GAImax (at anthesis) and GAD (Green Area Duration). Different regression models were compared: Yield = f(GAImax); f(GAD) ; f (GAImax, GAD). Linear, quadratic, linear-plateau, and quadratic with plateau models were tested. Models based on GAD or GAD + GAImax were among the best performing and the simplest ones to apply in practice. These models using remote sensing data at high temporal and spatial resolution provide a baseline for the comparison with process-based crop models such as SUNFLO (with or without GAI assimilation). Predicting oil production will be the next issue to address with this approach.

ABSTRACT. In conventional farming, fertilizers and crop-control substances are applied uniformly over fields. However, because the properties of the fields are not uniform, this approach leads to over-application in some places. To address this problem, precision agriculture (PA) was developed to rationalize inputs and reduce environmental impact.

This study investigates an imaging system based on a Rikola hyperspectral (HSI) and Nikon D800E (CIR) cameras installed on a manned ultralight aircraft Bekas Ch-32 for applications involving precision agriculture. The efficacy of this approach is compared with that of using Canon PowerShot SX260HS camera images acquired from helicopter-type unmanned aerial vehicles (UAVs) to accomplish similar tasks. Originally developed hyperspectral imaging systems provide images suitable for modelling chlorophyll concentration in spring wheat and for estimating the normalized difference red edge (NDRE) index, which is conventionally obtained using OptRx proximal sensors. Image values used as explanatory variables in ordinary least squares regression explain the variance in chlorophyll concentration and NDRE, respectively. Although hyperspectral images and CIR images may be acquired from ultralight aircraft and UAV, when used as the source of explanatory variables in ordinary least squares regression, the former leads to better predictions of chlorophyll concentration and NDRE than the latter.

ABSTRACT. Pan-sharpening methods have been developed to increase the spatial resolution of the multispectral information by fusing a panchromatic image (i.e. high spatial /low spectral resolution) with a multispectral one. In recent years, methods have been proposed for hyperspectral and multispectral data fusion.

These methods are generally used for satellite data with limited spatial resolution. In order to overcome the spatial resolution, sensors can be embedded on Unmanned Aerial Vehicles. In this presentation, we propose to study the impact of the spatial resolution of highly resolved spectral data on pan-sharpening methods in order to reconstruct a hyperspectral image and to choose the best combination of available cheap sensors.

ABSTRACT. Close range hyperspectral imagery is a promising tool for phenotyping or crop monitoring. In association with Partial Least Square Regression (PLS-R), it allows to build high spatial resolution maps of chemical content at the canopy level.

However, several optical phenomena must be taken into account when applying this approach to vegetation scenes in natural conditions. Among them, specular reflection and multiplicative factors due to leaf inclination can be overcome by pre-processing. But the most challenging phenomenon is multiple scattering: when a leaf receives part of its lighting from surrounding leaves, the incident light spectrum and thus the apparent reflectance are modified, leading to false regression results. Moreover, for practical reasons, it is impossible to include multiple scattering examples in a PLS-R calibration set.

We propose here an original method based on the analytic formulation of scattering, to be used in association with PLS-R. It only requires isolated leaf samples in its learning step. It has been developed using canopy and light propagation simulation tools, and evaluated on real crop plants for leaf nitrogen content assessment.

ABSTRACT. Plant protection products (PPP) have an impact on environment and human health. To adjust the spraying dose rate according to the crop to protect is a method to reduce the amount of PPP used in agriculture. Such a method requires defining relevant crop structure parameters. The present research was aimed to validate two crop structure indicators from a 2D Lidar at vineyard. The first indicator evaluated is the Tree Area Index (TAI). TAI is based on light interception probability. The second indicator has been named Leaf Wall Area by Point (LWApts). LWApts was a high resolution variant of classic Leaf Wall Area (LWA). Trials were conducted to compare and assess the two indicators for predicting actual spraying deposits of sprayers at vineyards. Two sprayers were used: a pneumatic arch and an air assisted face to face sprayer. Repeatability of Lidar indicators was also studied. Both indicators appeared repeatable and LWApts in particular. The predictions are very good for pneumatic sprayer and of a lesser quality for face to face sprayer. However, deposits are higher for face to face sprayer than for the pneumatic sprayer, which, in practice, may allow farmers to consider that higher efficiency allows for lesser dose rates

ABSTRACT. Japan is known to produce very high-value fruits such as famous white peach in Okayama prefecture. With a long hard work of the diligent farmers, advanced techniques and skills had been accumulated for fruit cultivation, such as pruning method, water and fertilizer management. Because those skills all require good understanding of the tree structures, and were not quantitively defined in real world dimension, it is currently very difficult for those skills to be inherited by the young farmers through just reading the paper based cultivation guidelines of the local institute. Therefore, the experts who are in charge of supporting young farmers are expecting a cultivation guidance system that can; (1) record the well managed peach trees in real world dimension, and (2) evaluate the tree phenotypic traits such as canopy structure (crown area, stem angle, etc.) quantitively based on 1. In order to built such system in the future, a combination of UAV-based field sensing and advanced image processing techniques (canopy segmentation from background which is poorly contrasted from tree canopy under the real field condition and isolation of the connected/overlapped canopies) are proposed.

ABSTRACT. Apple scab (ascomycete Venturia Inaqualis) is the main cause of stress and fruit losses in apple orchards. Its treatment is most often based on repeated fungicide applications, which is a costly and time consuming approach. In this context, early, accurate and non destructive detection of apple scab infection would be an efficient solution to optimize the management of the fruit disease, reducing fungicide applications while maintaining crop quality. Our study aims at exploring the potential of hyperspectral imaging for early detection of apple scab in leaves.

ABSTRACT. The goal of the study is to diagnose markets’ opportunities, over a 10 years projection, to see them in the light of research topics, in order to determine the innovation strategic axis and to help enhance technology and skill transfers.

The study has been steered by a committee of scientists from BRGM, CNES, CNRS, INERIS, INRA, INRIA, IRSTEA and Meteo France with inputs from CVT AllEnvi. This collective analysis has incorporated the findings from several market researches, ad hoc bibliometric studies and has completed them with numerous interviews of experts and two focused meetings, from both private and public sectors.

Main findings

There are technology transfer opportunities for precision agriculture on cross-cutting subjects taken up by AllEnvi researchers, such as soil moisture, the water cycle…

Transpose the methods developed for VHR images to moderate-resolution images.

Enhance partnerships on early detection of diseases and stress, farm inputs and phenology for arable lands, vineyards and vegetable crops;

Disseminate case studies to agricultural cooperatives and farmers about the benefits of inputs modulation;

Strengthen scientific and economic connections between the 3 observation scales: satellite, aerial and ground.

Technology transfer teams may use this study so they mobilize the best players on the selected markets.

ABSTRACT. A combined airborne thermal and multispectral imagery campaign was performed by UAV in an apple variety trial (flights at 40m elevation). The experimental orchard was dedicated to the assessment of various tree traits, including the response to water stress. Trees of each variety were either submitted to normal irrigation (WW), on the basis of soil hydric potential, or submitted (WS) to water restriction during summer period (July 2015). Measurements performed in this 'genotype by environment' experiment comprised a careful monitoring of fruit diametral growth, assessment of tree water status by stem of leaf hydric potential,while soil and climatic data were also collected. Thanks to remotely-sensed thermal imagery, significantly higher canopy temperatures were found in WS trees compared to WW ones, an also significant differences in red and NIR wavebands (multispectral images). Differences were linked to the severity of water deficit (leaf water potential). Responses also varied significantly according to the genotype: two varieties (more water spending ones) reached a characteristic stress while two others (more water saving) were less affected by water withholding, as revealed by image- and tree-based variables. The experiment thus demonstrates the potential of multispectral and thermal high resolution imagery to non-invasive phenotyping of fruit trees cover.

ABSTRACT. The adoption of small unmanned aerial systems endowed with cameras and other sensory systems has improved agricultural management practices. Most of the commercial platform available in the market are able to cover a set of way-points and trigger an individual action in each. However, the current mission planners just compute single mission. They fail to optimize the mission time and enable safety measures. This paper proposes the design of an aerial mission planner that overcome those issues. The aerial mission planner employs Ants Colony Optimization (ACO) algorithm to compute a complete aerial mission subject to a set of restrictions. The results shown that if the way-points are sorted in a specific manner the aerial mission can be optimized and the fleet safety preserved.

ABSTRACT. Aerial platforms such as planes and satellites are not suitable for a correct land use classification due to low spatial and temporal resolutions.Imagery based on unmanned aerial vehicles (UAVs) has become a feasible solution.UAV-methodology allows settings of flight and sensor parameters to optimize results.Many missions on the same location can be planned, which allow dynamical studies of temporal effects on land uses systems.In a typical flight with the USENSE-X8, an area of 2 km2 is covered at a resolution of 6.5 cm.The main objective of this research is the delineation and classification of land uses in the landscape.A considerable amount of image-based information is produced.As a result, an automated UAV data analysis workflow becomes necessary.In the pre-processing, UAV-imagery is filtered and stitched into an orthophoto and digital surface model based on structure-from-motion principles.The next step splits-up the data file into tiles to make parallelization of the computations achievable.A simple linear iterative clustering algorithm is applied to the orthophoto to extract superpixels.Individual superpixels are characterized by statistical descriptors and compared by classical statistical tests. A cluster analysis groups the superpixels in conceptual classes.Finally, ground truth is introduced and a discriminant analysis approach can be taken, resulting in misclassification estimates.

ABSTRACT. An experimental plot of oilseed rape was monitored via proxy and remote sensing tools during the vegetative/reproducing cycle. Non-destructive measurements included reflectance spectra (ASD FieldSpec), Field RGB images and RGB/Multispectral aerial images taken by UAV. Field observation and in situ measurements (biomass, LAI, N content, height) were simultaneously carried out on 1m² collected samples. Regarding the specific properties of the different growing stages, relevant measuring tools have been used for each stage. The field spectral measurements by ASD are highly correlated to the aerial data from UAV (NDVIUAV=1.003 * NDVIASD, R2=0.75 P<0.001). There was a strong correlation between a new image retrieved index and field/lab measurements (LAI/biomass) for similar vegetative stages (R²=0.87 P<0.001 n=105). The yellow index calculated from RGB images was a significant indicator of flower content. This index allowed establishment of flower rate classes in high accordance with field observations. Combination of several time consecutive images during the reproduction cycle can be used for evaluation of flowering procedure. There was also a strong correlation between seed yield and spectral index retrieved from multispectral image of green pods (R²=0.78 P<0.001 n=48) whereas the correlation between sampled biomass in the same date and seed yield was lower (R²=0.66 P<0.001 n=48).

ABSTRACT. Brazil has emerged as a global leader in the use of remote sensing data to monitor land use and cover dynamics with special focus in agriculture and forestry. This work aims to present some of the most important recent advances in monitoring the changes in land use and land cover and their interactions with processes of forest degradation in the Brazilian Amazon. The goal of this study is to evaluate land use trajectories and forest degradation processes. To quantify and understand the effects of degradation on forest structures, traditionally requires detailed work on forest inventory. We use active remote sensing through airborne LiDAR sensors (Light Detection and Ranging) combined with passive remote sensing using optical satellites. Among the main results and scientific, technological and innovation contributions of the study, we highlight: i) expansion of the LiDAR database and planning forest resources; ii) methodology for processing LiDAR data; iii) development of model for estimating carbon stocks in areas with intact, degraded and secondary forests; iv) advance in the knowledge of forest structure and the effect of degradation; and, v) new bases of technical and scientific knowledge in information and communication technology for natural resources planning and environmental management systems in Brazil.

ABSTRACT. Finding phenotype associations across multiple plant species, annotation strategies, and environments has become more difficult as the amount of annotated data has continued to increase. By associating annotations with ontologies as metadata, we can provide a structured, inferrable, and standardized context in which to improve our ability to mine data by more accurately defining our own data.
To this end, the Planteome project (http://planteome.org) ingests over 20 database sources, 80 taxa, and 2 million bioentities (genes, germplasm, QTL). Over 17 million bioentities are annotated to defined ontology terms in a standardized manner. With this infrastructure in place, Planteome provides a browsable resource for multiple reference ontologies for plants such as Plant Ontology (PO) describing anatomy and growth and developmental stages, Plant Trait Ontology (TO) describing phenotype traits, Gene Ontology (GO) describing molecular function, biological process and cellular components, Phenotype and Attribute Trait Ontology (PATO) and the Application ontologies that are species-specific Crop Ontology (CO). The database also allows for an ontology-based, faceted, cross-species search of plant phenomic and genomic data annotated with the reference ontologies. Data is denormalized using the GOlr infrastructure (http://wiki.geneontology.org/index.php/GOlr), built on top of the Solr search platform, providing quick and meaningful querying capabilities.
Work is currently underway to allow adopt a standardized Biolink web-services API (https://github.com/biolink/biolink-api) that, with GOlr, has already been adopted by the Monarch Initiative (https://monarchinitiative.org), an ontology-based tool for search and aggregation service focused on human disease through analysis of cross-species annotations.

ABSTRACT. Mid 2014, we started the AgroPortal project (http://agroportal.lirmm.fr) with the vision of offering a vocabulary & ontology repository for agronomy and related domains such as biodiversity, plant sciences and nutrition. The prototype found a good adoption, and growing interest appeared when presenting it to several interlocutors in the agronomy community (e.g., CGIAR (Bioversity International), INRA, IRD, CIRAD, IRSTEA, FAO, RDA, Planteome, EBI). We have now an advanced prototype platform which latest version (v1.3) was released in March 2017, that currently hosts 64 public ontologies including 38 not present in any such ontology repository (e.g., NCBO BioPortal) and 8 privates. This paper presents a short review of our current use cases and of the ontologies & vocabularies hosted in AgroPortal in Mai 2017. Thanks to a new ontology metadata model, we can now aggregate ontology descriptions to display information about the “landscape of agronomical ontologies” as presented.

ABSTRACT. Informational needs of agricultural consultants are increasingly complex. Advising farmers on the appropriate measures for optimizing cropping yields demands access to custom data archives and analytics tools. In line with the increasing number of archives, the expertise required of consultants is goes beyond the capabilities of these non-technical agri-specialists. These end users have diverse ad-hoc query needs and require tools that provide simple access to distributed data silos and easy ways to integrate relevant information. In this paper we report on a pilot deployment of SADI Semantic Web services for the federation and computation of agricultural data. A registry of 9 SADI services was deployed to expose data from a variety of different data resources in support of a defined set of query needs. We demonstrate that the deployment of these services facilitates the ad-hoc creation and execution of a number of a number of workflows targeting use cases in \textit{Agricultural Operations Management}. Using HYDRA, a semantic query engine for SADI with a custom built GUI, agricultural consultants can identify optimal crop varieties, compute profit margins of each variety using a complex cost model.

ABSTRACT. To ensure the usefulness of a bioinformatics service, analysis of user needs is an essential step. Furthermore, if the service anticipates the identified needs, acceptance by the user is easier. The aim of this work is to provide an overview of the requirements of a microbial diversity research community for ontology-based text-mining applications. This study is part of the development of the European infrastructure for text-mining, OpenMinTeD, that targets Biodiversity among other research fields.
The requirement analysis was completed through targeted online surveys, interviews, focus group meetings and workshops. This work yields to a detailed up-to-date landscape of stakeholders, their potential role and their expectations of general interest with respect to text-mining applications. Furthermore, we introduce a user-centered approach with the aim to produce more focused functional requirements, including application user interfaces. The resulting description of these needs guides OpenMinTeD current development to design and develop activities within text-mining projects for microbiology community.

ABSTRACT. High-quality big data is indispensable as training data to create agricultural applications such as plant growth models by machine learning. We propose architecture to create applications by massive sensor networks. The sensor networks consist of sensor nodes, which can be fabricated quickly based on open source hardware. The sensor nodes are low cost, multifunctional, durable, and easy to deploy in fields.

ABSTRACT. In 2010, the Brazilian Government launched the Low-Carbon Agriculture Plan to promote adoption of good management practices by farmers nationwide. The selected practices included zero-tillage and integrated crop-livestock systems. Remote sensing techniques have strong potential to measure the adoption of the practices. This paper aims at analyzing the spectral behavior of integrated crop-livestock systems compared to a forest and a degraded pasture.The study was conducted in the Embrapa Rice and Beans, Brazil. The climate had a two well-defined season. Ten paddocks were selected with areas ranging from 5.3 ha to 13.1 ha between April 2009 and October 2016. The crop rotation consisted of Brachiaria pasture grass in the dry season interchangeably with soybean, rainfed rice and maize+brachiaria pasture grass in the rainy season. We used 97 cloud-free NDVI images of Landsat-8 OLI and Landsat-7 ETM sensors for a 2009-2016 time series, acquired monthly, at 30 m spatial resolution. The results showed that sensor spectral response followed the vegetation growth. Sensors were also able to capture mainly the continuous high amount of forest biomass throughout the years with slight responses due to rainy and dry seasons. Sensors also captured the seasonal variations of crops and pastures that follow the rainy/dry seasons.

ABSTRACT. A recent trend in Europe has brought young farmers to settle on small surfaces, for example with vegetable crops. A major challenge on such microfarms is to overcome the workload as the small surfaces are usually compensated for by using intensive culture techniques that require a large amount of manual work. The traditional machines that are used in agriculture are usually optimized for speed and power and not adapted to dense cultures. The LettuceThink platform is a lightweight wheeled robot equipped with precision sensors and actuators controlled with open source software. Below, we present algorithms for two applications relevant to market farms. The weeding of crop beds is the task that demands a lot of work and it is thus the first application that we wish to demonstrate. As a second application, we present the precise characterization of plant growth and plant structure, which provides useful information to prevent the spread of diseases and to plan harvests.

ABSTRACT. An characteristic of Brazilian livestock is the grazing system, mainly because it is the most economical and efficient way of producing and supplying feed for cattle. Due to this factor, Brazil has one of the lowest costs of meat production in the world; however, the potential productivity of livestock farming can be reduced when developed extensively. Thus, one of the great challenges faced in livestock production is the increase of efficiency and, consequently, the yield of the productive system by using of technologies and the development of monitoring systems that possibility the technological advance of the sector. There are several applications in the livestock industry for the use of electronic identifiers, but the broad adoption is repressed due to the high costs of existing equipment. This project aims to develop a low-cost alternative to monitor the simultaneous visit of animals in collective troughs, in order to measure physiological, behavioral and productive indicators of the animals under pasture conditions. The equipment demonstrated functionality and it was possible to identify that the construction of this station presents a competitive cost, as well as allowing the flexibility to add new functions of information capture, through other sensors and actuators.

ABSTRACT. In vineyards, the scouting carry out for the disease detection are a high time consuming operations, because a specialist must assess every plants searching the presence of illness symptoms. Goal of this research was to develop and test a system suitable to carry out crop monitoring tasks, with the aim to make faster these operations. The proposed system consider the application of different techniques: remote sensing, for a fast recognition of critical areas on large scale; a ground-based sensing to identify the diseased plants in those areas and a visual survey to investigate in detail the symptoms. The obtained results highlighted an appreciate capability of the system to recognize critical areas in the vineyard as well as to recognize the presence of low vigor conditions along the vine rows. Considering the comparison between the remote and ground-based sensing with the visual survey, the UAV and ByeLab surveys were able to recognize the 88% of low vigor conditions states. Due to the satisfactory results the proposed system could be a good solution to carry out faster crop monitoring operation, indeed the specialist can focus his attention directly to the plants that show disease symptoms avoiding the scouting of the whole vineyard.

ABSTRACT. Electronic services of information, consultation and training in implementing integrated plant protection IKMIS implement the means to provide 4 electronic services, which are acces-sible openly for public and are free of charge: (1) categorised provision of information on the distribu-tion of harmful organisms in crops all through Lithuania; (2) provision of informative electronic cata-logues for farmers about diseases, pests and weeds as well as plant protection products registered in the country; (3) provision of interactive consultation services regarding the issues of integrated plant protec-tion; (4) organization of remote training for farmers, agricultural specialists and advisers including ac-creditation trainings and issuance of certificates. The electronic services delivered through the system that are available at www.ikmis.lt or via VIISP sys-tem Lithuanian e-Government Gateway or via the portal designat-ed for farmers www.agroakademija.lt. Public Institution Lithuanian Agricultural Advisory Service (hereinafter referred to as LAAS) has devel-oped IKMIS taking into account European agricultural issues relevant today while implementing the EU strategy for sustainable pesticide use, the Plant Protection Law of RL and other legal acts. The project IKMIS has been developed in cooperation with the Lithuanian Research Centre for Agricul-ture and Forestry, Aleksandras Stulginskis University and the Centre for LEADER Programme and Agri-cultural Training Methodology.

ABSTRACT. Many software systems in the field of agriculture exists today, provided by several vendors, developed for different purposes. As part of them, farm management information systems differ in their use (field catalog, herd managers, etc.), capabilities (e.g. balances, analysis) and distribution (local PC on farm sight, Web Applications and hybrids). However, they have got on thing in common: The focus is the farm or parts of it. In the view of ICT’s, a farm is defined through its data (e.g. fields, farm animals, activities) located in time and space. The scope of data parameters is not a subject of frequent changes. The data which is acquired by farms is mostly what have been done, when, where, who and the economical information. What currently changes is the amount of the data. This is caused by the rapid development of new technical devices like smart devices and sensors (e.g. measuring nutrient content in crops for fertilizations). Currently every solution has its own internal representation of the farm data. The function must be designed and implemented by every vendor. This contribution describes a farm software ecosystem with a base architecture and its core components to build an open source software.

ABSTRACT. The communication addresses the characteristics of innovation processes regarding the new technologies of information and communication and the role of research to contribute to such innovation processes. It highlights how important it is to take into account the users in the conception of the technologies and to think in terms of co-design of the technologies. It also shows that public or private research could be based on different models to contribute to the conception and diffusion of these technologies.

ABSTRACT. Over the past few years, important investments have been devoted to the development of automatic field high throughput phenotyping systems (FHTPS), as part of projects like Phénome. Such FHTPS embed specific sensors (spectroradiometers, RGB cameras, LIDARs) and include softwares that allow the computation of vegetation biophysical parameters. Information flows coming from those FHTPS have a high spatial and/or temporal density. Furthermore, measurements must be combined with metadata in order to make them understandable and integrate them into our internal historic database of trials (IHDBT). Besides, they have to be quickly and easily available to the user for validation and study. The AgroPhenX information system was built to achieve these purposes: (i) management of high volume of measurements, (ii) traceability of their provenance and (iii) easiness for user access. Measurements are user-viewable through a web interface which is an useful medium for outlier detection and a better adoption of high throughput phenotyping measurements by the user. Thanks to an electronic data interchange system, AgroPhenX and the IHBT are interoperable: the user can easily import data coming from FHTPS in the IHBT. The high volume of the FHTPS data required several optimizations of AgroPhenX, such as the implementation of materialized views.

ABSTRACT. The objective of this study is to greatly reduce the labor involved with plant shape measurement for trait evaluation in breeding research, especially in large-scale outdoor fields, by using image processing. Our primary focus is data acquisition and data management for plant phenotyping. Regarding data acquisition, as a method for easily measuring plant shapes in the field, we use ortho-mosaic and three-dimensional (3D) reconstruction data of the plant shape based on continuous shooting of the field over time. Regarding data management, our final goal is to provide an analysis platform that links genotypes and phenotypes and to manage them in such a way that breeding researchers can easily analyze them. We are currently developing a phenotype analysis platform that can consistently provide the functions necessary for field phenotyping from data acquisition to analysis, as a milestone of the final goal. In the present work, we propose a phenotype analysis platform that measures and aggregates phenotypes automatically using ortho-mosaic and 3D reconstruction data generated from farm field images.

ABSTRACT. For High-throughput phenotyping by using Agri-big data on field level, it will be expected to open the next door of plant breeding. It is important first step to capture correctly for leaf elongation and expansion pattern and speed in early stage. It would strongly connect yield performance. UAV technology can easily provide us as numerous images of growth record of plants in a field. So, we tried to reconstruct as 3D image by using digital image analysis captured image files. We investigated for canopy height for the relationship of all F1s. We named “4D score” was admitted apparent negative correlation between root weight and 4D_score. We concluded a new concept of 4D_score could be able to suggest valuable scope of sugar beet Heterosis as numerical value.

ABSTRACT. The main objective of providing digital data via web platforms is to facilitate sharing and exchanges between actors, and to enable the enrichment of data by its various users. The ability of computer systems to function with other systems or products, defines the concept of interoperability. To do this, interoperability should refer to a standard aimed at structuring the data so that it is understood by the different actors and systems.

The CIRAD had set up a web cartographic platform, called AWARE, gathering all the geographical data available and produced by researchers based in Reunion Island. Technically, AWARE is based on GeoNode, an open source geospatial content management system and follows the INfrastructure for SPatial Information in Europe directive (INSPIRE) in order to be compatible with French geographical data producers. INSPIRE thus ensures semantic interoperability.

GeoNode ensures technical interoperability by implementing the Open Geospatial Consortium (OGC) better known as web services. This interoperability allows AWARE to be harvested and to harvest other remote geographic catalogs.

By creating the AWARE platform, the ARTISTS team has created a true digital heritage of spatial information for CIRAD in La Reunion Island.

ABSTRACT. Recently, grain legumes have been in the spotlight of agronomic research since their introduction in cropping systems is considered as one of the levers available for the transition to more sustainable agrofood systems. Unfortunately, studies on grain legumes identified several socio-technical lock-in currently limiting their integration in French agriculture, including a serious lack of technical knowledge and references available for the farmers.

The purpose of this study was to evaluate the utility of two operational tools developed by the French organisation Club Gestion des Connaissances, mainly composed of industrial and services sectors members, for the establishment of a preliminary knowledge management diagnosis in an agricultural domain. To this end, we conducted a survey in six major French lentil production areas. The first tool we used, called Critical Knowledge Factors, is a questionnaire rating the criticality of the different fields of knowledge associated to a particular practice. The second tool we used, called Community Maturity Model, is a questionnaire evaluating the maturity of a community of practice in terms of knowledge management. The results of this survey prove that those two tools are useful for analysing knowledge management practices and needs in an agricultural field.

ABSTRACT. In recent agricultural field, artificial intelligence and big data analysis technologies are expected. It is important but difficult to collect large amount of various agricultural data (AgriBigData). In our previous study, we have developed an agricultural sensor network named Field Server. In this study, we have proposed and developed data harvesting system for AgriBigData based on Field Server technology. The data harvesting system consisting of a swarm of drones, stationary Field Servers, and a robotic Field Server. The drones collect crop community image from top view. The stationary Field Servers, modified for non-expert users to obtain, deploy, and use them easily, monitor environment condition and crop growth condition clearly from another point of view with high performance cameras. The robotic Field Server, developed with a six-legged platform and a lightweight manipulator, monitors target crops closely or from within the community. Using a combination of these devices, we can construct useful AgriBigData by covering disadvantages of each monitoring. To evaluate our proposed system, we conducted some experiments to collect AgriBigData with the system, and we have successfully collected more than 200 GB data. Throughout the experiments, we can demonstrate our system’s effectiveness and potential, and clarified future works.

ABSTRACT. Creeping thistle is a perennial weed that tends to grow in patches and causes significant yield loss in cereal crops. A tool for detecting thistles, Thistle-Tool, have been developed by Rasmussen et al. The tool detects the level of green in images and compares it with a treshold. The images are taken by a drone in altitudes of 10m and 50m, divided into sub-images covering 1m^2 and feed to the model. This project aims to use Convolutional Neural Networks (CNNs) to show that the results achieved with Thistle-Tool can be improved. The network chosen for the model is based on DenseNet, a state of the art CNN within image classification. The images are annotated as crop with or without thistles. Thistle-Tool achieves an accuracy of 96% in winter wheat and 80% in spring barley, while the CNN achieves more than 95% on both crops combined. This shows that by using CNNs it is possible to improve the results of current version of Thistle-Tool. Advantages of using the CNN is that no threshold is needed and that features are found automatically by training. This method may possibly be transferred to other similar detection related cases if annotated data are available.

ABSTRACT. Whiteflies have two major biotypes, B-biotype and Q-biotype, and identification of biotypes are necessary for pest controlling, since they have different pesticide resistance. However, identification of biotypes requires slow and expensive techniques in exchange for accurate decisions. In this paper, biotype identification of whiteflies using acoustic signature is proposed, and its performance is evaluated in experiments. The proposed scheme achieves biotype identification by three steps; 1) automatically detection of sound of whiteflies, 2) analysis of detected signal in frequency domain, and 3) identification of biotypes using acoustic signature. Performance of the proposed system was evaluated in experiments. Sound of B-biotype and Q2-biotype was recorded, automatic detection, signal analysis, and calculation of inner product were performed. As a result, it was found that can identify biotypes because inner product values of sound and template that belong to the same biotype are large, the different biotype are small. Biotype identification of whiteflies using acoustic signature was proposed, and its performance was evaluated in experiments. The proposed scheme achieved biotype identification by using acoustic signature. Performance of the proposed system was evaluated in experiments, and the obtained results suggest that the proposed scheme is cost-effective alternative for biotype identification.

ABSTRACT. The contribution presents a concept study of a robotic approach to supply single plants on arable land with a plant-specific amount of fertilizer on a daily, weekly or other time basis. The study is being developed in an interdisciplinary research project integrating aspects from plant science, technology and farm economy. We will describe the conceptual robots and the fertilizing operation in detail and present first results of an agent-based operation simulation. One output of the simulation is for instance the required number of robots for a given amount of fertilizer and a given field size and form. A high-frequency fertilizer application with low amounts of fertilizer per application is expected to have ecological benefits due to reduction of fertilizer run-off as well as economic benefits due to increased fertilizer efficiency. We will give a first economic evaluation of our concept. The developed concept requires certain technological frameworks, such as a communication link between fields and management servers or field stations for refilling and recharging operations. These requirements will be defined by our contribution.

ABSTRACT. Precision agriculture generally requires collecting data through regular sampling or sensor maintenance in the field. Managing technicians who perform these tasks becomes very complex when the number of plots increases. This research was motivated by a real problem of technician scheduling. Fruition Sciences is a information technology company that provides a web-application to winemakers and grapegrowers to optimize grape quality and yield. The Technician Scheduling Problem presented here consists in finding a planning of technicians that satisfies all constraints and minimizes the travel time of all technicians. We propose a preliminary approach based on Constraint Programming to solve the issue of the problem.

ABSTRACT. This paper describes our general procedure to support a decision for agro-waste management analysis. The problem we are considering is how to make a “good” decision regarding issues coming from agricultural engineering with the aid of Computational Social Choice (CSC) and Argumentation Framework (AF). Even though CSC and AF have been used autonomously to support decision-making we believe that by combining these two fields we can propose social fair decisions by taking into account both (1) the stakeholders preferences and (2) the justifications behind these preferences. Therefore we implement a software tool for decision-making which is composed by two main modules, i.e., the social choice module and the deliberation module. In this paper we describe thoroughly our tool and how it can be applied to different alternatives on the valorisation of agro-waste. As an example we present an application of our tool in the context of IFV’s survey where the decision to be taken is to derive the best solution for managing the oenological by-products.

ABSTRACT. The increasing development of information and communication technologies (ICT) in agriculture has open the way for changes in the production and dissemination of knowledge. However, this knowledge economy is faced technic, social and economic stakes. From an economic perspective, digital development involves a transformation of farm’s business models and constitutes a source of value and competitiveness. This paper questions the interdependencies between the development of digital devices and the productivity gains in farms. A microeconomic methodology based on productivity analysis was used to compare economic performance of some dairy farms surveyed and economic performance of the average upper quarters of dairy farms in west of France. Results have permitted to identify relationship between digital technology and operations affected the dairy farms and have shown that in dairy farms, technology like milking robot system impact positively the milk production, labor organization, animal feedstuffs and the overall quality of herd management. However these impacts remain variable according to the nature of the used technology and to the farmer experience. Results shows also those farms equipped with digital tools are more productive when these tools are fully integrated as a part of a rather homogeneous system, since they are complementary in their action.

ABSTRACT. Food security is important for a healthy nation. Agriculture is an important sector for Kenya’s economic growth and for achieving Kenya’s Vision 2030. However, change in farming practices from one season to the other, increased demand for land, intensive land use and failure to apply recommended farming practice, has resulted in soil degradation over time. Land evaluation is the process of assessment of land performance for specified purposes e.g. crop production. Currently, in the Department of Soil Survey at Kenya Agricultural and Livestock Research Organization, and in many other soil survey institutions, land evaluation process is done manually and is time consuming, stressful and prone to human errors. We review computer science contributions in land evaluation process and other areas in agriculture. Parallel Random Forest (PRF) was identified to be performing better that other Machine Learning Classification algorithms. An experiment prototype was set up and PRF gave significant scores in Root Mean Square Error, Accuracy and Speedup. This prototype can be scaled up in precision farming, to optimize prediction of suitable crop(s) from soil samples information. This will improve the speed and accuracy of dissemination of the knowledge and ultimately improving crop and livestock productivity.

ABSTRACT. In the Netherlands the soil nitrogen supply and the crop nitrogen requirement varies widely from year to year and from field to field. For potato farmers sidedress systems based on physical measurements are available, however these systems neglect the spatial variation within fields. Based on 20 years of research we developed a crop reflectance based N sidedress system for potatoes which overcomes this problem. We implemented this system in a web-based portal called Akkerweb, which allows for safe and easy storage of spatial and temporal soil, crop, climate and management data. Since 2014 we made a small group of farmers acquainted with this application. In participative projects we organized an entire workflow from measuring with tractor mounted sensors (a Yara-N-Sensor) to a N sidedress recommendation. In 2016 field experiments were done to fine-tune the recommendation system, to enable recommendations on the basis of UAV-images and to estimate the usefulness of several other near-by sensors. For 2017 we plan to improve our model-based recommendation by implementing a crop growth model in combination with using the gathered field data from the user-group.

ABSTRACT. Lignocellulosic biomass and agro-waste valorization routes are two of the promising methods towards a more sustainable bio-economy. Scientific literature in this domain is increasing fast and could be a valuable source of data. As these abundant scientific data are mostly in textual format and heterogeneously structured, using them to compute biomass treatment efficiency is not straightforward. The implementation of a Decision Support System (DSS) based on an original pipeline coupling knowledge management (KM) based on semantic web technologies, soft computing techniques and environmental factor computation has been done for lignocellulosic biomass valorization routes into glucose. The DSS allows using data found in the literature to assess environmental sustainability of biorefinery systems. The pipeline permits to: (1) structure and integrate relevant experimental data, (2) assess data source reliability, (3) compute and visualize green indicators taking into account data imprecision and source reliability. This pipeline has been made possible thanks to innovative researches in the coupling of ontologies, uncertainty management and propagation. In this first version, data acquisition is done by experts and facilitated by a Termino-Ontological Resource (TOR) called Biorefinery. The communication is dedicated to Biorefinery TOR model.

ABSTRACT. Food chains consist of many links and operate on a global scale with many stakeholders involved from farm to fork. Each stakeholder maintains data about food products that they handle, but this data is not transparently available to all stakeholders in the chain and trust in data sharing is low. In addition, there are various other data sources that contain interesting data for stakeholders in the food chain, such as import/export transactions, production (forecast) data, parcel crop information, local weather predictions and social media streams.

Combining data sources allows for (big) data analysis, pattern searching and thus better decision-making on when to produce what quantities of which food products. These data sources can be made easily accessible via a semantic web-based mechanism in which secure, linked data principles are applied. To achieve this, the Dutch horticulture and food domain is developing the HortiCube platform via which various data sources are made available to application developers using a general linked data application interface.

ABSTRACT. Nowadays a new concept is emerging to influence the evolution of existing Agriculture information system design: the data lake concept. This new concept is more a data driven information system than an information driven systems. What is this new concept? What is its definition? What is it composed of? What is its position in relation to the traditional data warehousing / analytics architecture? What are the main components of the architecture? How will the design of information systems for agriculture be influenced? For this domain, it is crucial to explore all data sets available and accessible (such as spatial data set) structured but also non-structured in order to find new knowledge and insights. The data lake is going to be positioned as a data capitalization systems for all type of data, in their raw format which is going to act as a information incubator to finally populate the existing decision support system. In this submission, we propose a new approach for getting new insight, knowledge and information for agriculture and agriculture practices, the data lake approach.

ABSTRACT. In this paper, we describe the landscape of data standards of relevance to the agrifood sector including all stages from research, through food production up to and including retail and consumers. The research reported here is part of a wider attempt at developing a strategic analysis of the standards, in part to identify gaps and overlapping standards, but also to provide an evaluation (both objective and subjective) of the quality and utility of the standards available. We explain how with the growth of the current generation of information technology and the development of data driven innovation in the agrifood sector, there is a fundamental need for a more joined up approach to the integration of the agrifood sector, its data streams and consequently the data standards which underpin these data flows. We argue that if effective use is to be made of data developed at the research stage for new agricultural products (seeds, crop varieties), then this has to be integrated with the food production stage for ongoing feedback. Equally data from the farm to the final consumer needs to be able to be integrated and with such a variety of incompatible data standards this currently is excessively difficult.

ABSTRACT. The agricultural system has experienced a strong abandonment of agro forestry in the 20th century, to count today only a few million ha in Europe. One of the aims of the AgroFE and the Agrof-MM European project to build a knowledge databank to help the education development, experts, farmers with agroforestry knowledge. The KDB (Knowledge DataBank) is based on different professional vocabulary, metadata and thesaurus system what are used for building the content structure and helping the users in searching. A new tool will improve knowledge management of agroforestry. The Agrof-MM project is creating an agroforestry thesaurus.