A High-Throughput Phenotyping Robot for Measuring Stalk Diameters of Maize Crops

High-ThroughPut Phenotyping (HTPP) is currently an active research topic in crop breeding. However, field-based HTPP still suffers from certain bottlenecks, such as small operating space and strong light variation of application scenarios. To address this problem, this paper proposes a solution for in-field HTPP. First, we develop an ultra-narrow phenotyping robot platform that can travel in-row and under canopy. Then, we deploy the Convolutional Neural Network (CNN) on our robot system to detect the maize stalks. Finally, we present an approach to calculate the stalk diameters based on RGB-D camera data. Here, CNN is used to detect maize stalks in RGB images, while the depth images are used to calculate the widths of the stalk bounding boxes, which are considered as the stalk diameters. The field experiment results show that the maximum deviation of stalk diameters measured by our approach is 0.007 m, RMSE is 0.003. To sum up, our robot system can be successfully applied in field HTPP.