CC BY 4.0Engler, HannesHannesEnglerGauweiler, PascalPascalGauweilerHuber, FlorianFlorianHuberKrause, JuliusJuliusKrauseFischer, BenediktBenediktFischerHoffmann, BenediktBenediktHoffmannSchumacher, PetraPetraSchumacherYushchenko, ArtemArtemYushchenkoGruna, RobinRobinGrunaSteinhage, VolkerVolkerSteinhageHerzog, KatjaKatjaHerzogTöpfer, ReinhardReinhardTöpferKicherer, AnnaAnnaKicherer2024-01-162024-01-162023https://publica.fraunhofer.de/handle/publica/458866https://doi.org/10.24406/publica-244210.5073/vitis.2023.62.special-issue.41-4810.24406/publica-2442Balanced and stable yield is a major trait in grapevine breeding and breeding research. Grapevine yield hereby is a complex quantitative trait, as it is influenced by multiple plant parameters, like berry size, number of berries per bunch, number of bunches per shoot, management, and environmental factors. In the current breeding process, the complexity of this trait has shown that a classification according to descriptive factors for marker development is only possible to a limited extent. Precise field phenotyping of yield-related traits is the basic prerequisite to be able to measure such quantitative traits. This, however, is the major bottleneck due to labor, time and constrains of plant material in the breeding process. For this reason, one of our main goals with the newly developed phenotyping platform PHENOquad with its multisensor system PHENOboxx is to improve phenotyping efficiency of grapevine yield to overcome the phenotyping bottleneck. The newly developed embedded vision system PHENOboxx is mounted on an "all-terrain vehicle (ATV)". This allows a fast data acquisition on a large number of individual vines. In order to evaluate the yield potential of breeding material in comparison to established grapevine cultivars, various yield-related parameters of the vines are quantified directly in the field with high spatial and temporal resolution. As key parameters for yield-related phenotyping, the number of shoots, bunches, berries and the weight of dormant pruning wood was identified. The image data acquired are annotated to train the artificial intelligence (AI). Within the process, the image analysis results are compared to annotated ground truth data and correlated with the field reference data. We expect to increase the precision, target specificity and throughput of screening grapevine material without reducing its accuracy over time by using the PHENOquad. In addition, a weighting of yield-relevant parameters would be possible. This opens up new possibilities for efficient plant evaluation in the scope of grapevine breeding. Also new application possibilities for precision viticulture are conceivable.enPhenolinerPHENOboxxHigh-throughputphenotypingprecision viticulturesensordata managementimage analysisyield componentsjournal article