Files
Abstract
Agriculture is facing tremendous challenges to meet the needs of a growing world population which is projected to exceed 9 billion by 2050. It is urgent to build a sustainable agriculture system in order to accelerate crop improvement while reducing the environmental footprint. High through phenotyping, which refers to the process of measuring and assessing qualitative and quantitative plant bio-physical traits in an efficient and effective manner, is instrumental both in facilitating new plant breeding technologies and improving crop management practices.
This dissertation mainly focused on the development of field-based 3D imaging systems and methodologies for plant trait mapping from canopy level to organ level. Cotton plants were used as the model plants since cotton is among the most economically important crops providing natural fiber throughout the world. A ground vehicle-based 3D plant canopy surface reconstruction system, mainly consisting of a 2D line scan light detection and ranging (LiDAR) sensor and a real-time kinematic GPS, was developed to scan plants periodically. Algorithms were developed to extract three morphological traits at plot level and monitor their growth rate. In order to reduce the occlusion problem which was a major reason for the underestimation of cotton boll counting from images, a multi-view camera system was developed to scan plants from different perspectives. Then, dense point clouds were reconstructed from the images using structure from motion algorithm, from which a supervoxel-based segmentation and density-based clustering algorithm was developed to map cotton bolls. In addition to boll number, its size and position were extracted. A data processing pipeline, including skeleton extraction, main stalk and individual branches segmentation, and node localization, was developed to map plant nodes from point clouds obtained using a high-resolution 3D LiDAR. Experimental results showed that the developed systems and methodologies can accurately and efficiently measure these phenotypic traits. Furthermore, in addition to automating the types of the traits measured manually, the developed systems and methodologies can measure new phenotypic traits. Therefore, they can be used as tools both for plant researchers and growers alike, and can be used for other plants such as wheat after minor modifications.
This dissertation mainly focused on the development of field-based 3D imaging systems and methodologies for plant trait mapping from canopy level to organ level. Cotton plants were used as the model plants since cotton is among the most economically important crops providing natural fiber throughout the world. A ground vehicle-based 3D plant canopy surface reconstruction system, mainly consisting of a 2D line scan light detection and ranging (LiDAR) sensor and a real-time kinematic GPS, was developed to scan plants periodically. Algorithms were developed to extract three morphological traits at plot level and monitor their growth rate. In order to reduce the occlusion problem which was a major reason for the underestimation of cotton boll counting from images, a multi-view camera system was developed to scan plants from different perspectives. Then, dense point clouds were reconstructed from the images using structure from motion algorithm, from which a supervoxel-based segmentation and density-based clustering algorithm was developed to map cotton bolls. In addition to boll number, its size and position were extracted. A data processing pipeline, including skeleton extraction, main stalk and individual branches segmentation, and node localization, was developed to map plant nodes from point clouds obtained using a high-resolution 3D LiDAR. Experimental results showed that the developed systems and methodologies can accurately and efficiently measure these phenotypic traits. Furthermore, in addition to automating the types of the traits measured manually, the developed systems and methodologies can measure new phenotypic traits. Therefore, they can be used as tools both for plant researchers and growers alike, and can be used for other plants such as wheat after minor modifications.