Efficient Sensing for Long-Term Crop Monitoring

用于长期作物监测的高效传感

• 批准号: 498565298
• 负责人: Professorin Dr. Maren Bennewitz
• 金额: --
• 依托单位: Arbeitsgruppe für Humanoide Roboter
• 依托单位国家: 德国
• 项目类别: Research Units
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/498565298?language=en
• 关键词: Efficient; Sensing; Long; Term; Crop

Long-term monitoring of horticultural crops is necessary to realize the forecasting of phenotypes and enable good management decisions. Autonomously, learning 3D models of crops is difficult due to their complex structure, occlusions that naturally occur, and deformation, or growth over time. Typically, multiple viewpoints are required to obtain relevant information of a crop, e.g., the position and shape of flowers, fruits, and the stem; furthermore, the sensing has to be carried out frequently to improve the phenotype forecast. Therefore, we will develop active sensing strategies to measure crops over their life cycle. We will provide methods to enable efficient sensing of relevant features and improve state-of-the-art next-best-view planners. We will make viewpoint planning efficient by considering the required regions of interest, which contain the relevant features for decision making. Furthermore, we will exploit prior information to speed up repeated coverage over time. We will use information about previously perceived crop parts and newly gained knowledge about significant features to speed up the coverage of relevant parts. Finally, we will optimize the sequence of the necessary viewpoints by trading off the cost of sensor data acquisition and the resulting reduced forecast uncertainty. To summarize, project IP3 will provide methods for efficiently obtaining relevant sensor data of horticultural crops and enable efficient decision making for necessary management actions in the other IPs.

对园艺作物的长期监测对于实现表型预测和良好的管理决策是必要的。自主性地学习作物的3D模型是困难的，因为它们的复杂结构、自然发生的遮挡以及随着时间的推移而变形或生长。通常，需要多个视角来获得作物的相关信息，例如花、果实和茎的位置和形状；此外，为了提高表型预测，必须频繁地进行传感。因此，我们将开发主动传感策略来测量作物的整个生命周期。我们将提供方法，以实现对相关特征的有效感知，并改进最先进的次佳景观规划师。我们将通过考虑所需的感兴趣区域来提高视点规划的效率，这些区域包含用于决策的相关特征。此外，我们将利用先前的信息，随着时间的推移加快重复覆盖的速度。我们将利用以前感知到的作物部分的信息和新获得的重要特征知识来加快相关部分的覆盖。最后，我们将通过权衡传感器数据采集的成本和由此减少的预测不确定性来优化必要观点的顺序。综上所述，IP3项目将提供有效获取园艺作物相关传感器数据的方法，并使其他IP能够有效地做出必要的管理行动的决策。