京都大学 / 0048 / 新制・課程博士 / 博士(農学) / 甲第21824号 / 農博第2337号 / 新制||農||1067(附属図書館) / 学位論文||H31||N5196(農学部図書室) / 京都大学大学院農学研究科森林科学専攻 / (主査)教授 大澤 晃, 教授 北山 兼弘, 教授 北島 薫 / 学位規則第4条第1項該当 / Doctor of Agricultural Science / Kyoto University / DGAM

<p dir="ltr">High-throughput precision agriculture is revolutionizing farming by integrating advanced technologies to enhance productivity, efficiency, and sustainability. Tools such as three-dimensional (3D) imaging, remote sensing, and data analytics enable rapid collection and processing of large data volumes, allowing farmers to make informed decisions about crop management, soil health, and resource utilization. These technologies reduce waste, minimize environmental impact, and optimize inputs like water and fertilizers, addressing global food demand and resilience to climate change. Among these tools, high-resolution 3D imaging and close-range hyperspectral and multispectral imaging are particularly valuable. High-resolution 3D imaging provides detailed spatial data on plant structures, essential for monitoring growth, drought response, and phenotyping. Meanwhile, hyperspectral and multispectral imaging captures a broad spectrum of light, revealing biochemical and physiological details, including early signs of disease and nutrient deficiencies.</p><p dir="ltr">However, applying these technologies in high-throughput precision agriculture presents challenges. Structured light phase-shifting method, despite its accuracy in 3D shape measurement, is often limited by the time-consuming requirement of multiple projected patterns for a single 3D reconstruction. To address this, we developed a method that reduces the number of required patterns by incorporating a single additional binary random pattern, combined with digital image correlation and phase-shifting, enabling high-speed 3D shape measurement. Further, we embedded this additional pattern into the fringe patterns and utilized the Semi-Global Matching algorithm to enhance measurement speed and assist in absolute phase acquisition for 3D reconstruction.</p><p dir="ltr">Similarly, while hyperspectral and multispectral imaging provide rich spectral detail, they are often compromised by illumination variations due to complex specimen geometries. To overcome this, we propose an image-based method utilizing high-resolution 3D imaging for calibrating non-isotropic point light sources, addressing challenges posed by their non-radially symmetric radiant intensity distribution and the non-Lambertian properties of the calibration target. Our experiments demonstrated the high accuracy of this calibrated light source model, which was successfully applied to estimate the relative reflectance of various surfaces.</p><p dir="ltr">Finally, we demonstrate the effectiveness of high-speed, high-resolution 3D imaging combined with multispectral imaging for leaf vein segmentation and density measurement, where venation traits serve as critical indicators of plant physiological function. We developed a multimodal imaging system capable of capturing grayscale, multispectral, and 3D data in a unified coordinate system. Additionally, we designed a leaf vein segmentation framework that leverages all three imaging modalities to test the hypothesis that integrating 3D imaging can improve vein segmentation and density measurement without the need for labor-intensive leaf clearing or destructive 2D flatbed scanning. Our findings indicate that 3D imaging provides valuable topographical information, enhancing vein segmentation, albeit with some false detections. Additionally, this approach offers clear advantages in directly measuring morphological traits, such as leaf area, vein length, and vein density, especially in unstructured environments where traditional methods often fall short.</p><p dir="ltr">In summary, this dissertation advances the fields of high-speed, high-resolution 3D imaging and multispectral imaging in precision agriculture by: (1) reducing the number of patterns required for single-frame 3D reconstruction through the integration of digital image correlation and the Semi-Global Matching algorithm with phase-shifting techniques; (2) developing an accurate non-isotropic point light source model to correct for illumination variations in multispectral images; and (3) enhancing leaf vein segmentation and vein density measurement in non-flattened sugar maple leaves using high-resolution 3D imaging. These contributions address current challenges, providing robust solutions that pave the way for more efficient and sustainable agricultural practices.</p>