#Mobile Laser Scanning

提出基于关键点检测的甘蔗田间杂草生长点定位方法，为激光除草提供精准靶标。该方法在复杂田间背景下实现生长点鲁棒定位，支撑自动化精准除草作业。

开发低成本RGB与点云融合扫描系统，用于小麦表型测量。

提出多光谱反射场框架，实现近距植物三维点云辐射精确生成。

对比无人机RGB与LiDAR传感器在甘蔗田地形测绘中的精度与适用性。

结合SfM、LiDAR和TLS监测芬兰北部冻胀丘的年度表面变化与退化动态。

系统综述工业环境中计算机视觉在检测、安全、物流与机器人四领域的应用现状与挑战。

稻田定点杂草管理相比传统广域喷药可大幅减少除草剂用量,但现场部署受三大挑战制约:冠层下GNSS退化的稳健作物行导航、稻与形态多样化杂草的实时视觉区分、将稻误识为杂草的不可逆错误代价。本研究提出AgriNav,围绕四个ROS耦合模块构建集成自主拖拉机系统:用于稻检测的WeedDet PyTorch重实现、并行轻量级168万参数CNN-FPN变体(带非对称类加权)、通过硬编码置信门否决保护稻类的逆逻辑判别模块,以及融合GNSS/IMU/轮式里程计的6状态恒定速度转向率扩展卡尔曼滤波。主要系统级贡献是LiDAR-相机融合桥,利用导航LiDAR做ROI约束、世界坐标投影、地面平面滤波和双向置信融合,无额外硬件成本。

Abstract Context Spatially explicit and accurate maize crop height maps are essential for supporting site-specific management in precision agriculture. Although low-cost unmanned aerial system (UAS)-based multispectral imagery offers a promising approach for maize crop height estimation, it’s accuracy under varying canopy conditions remains uncertain. Aims In this study, we used 3D point clouds to estimate maize crop height during sparse and dense canopy cover based on UAS-captured multispectral images and compare with light detection and ranging (LiDAR)-based estimates. Methods The 3D point clouds from multispectral images and LiDAR data were processed using a spatially localized grid (SLG) filtering approachto derive digital terrain model (DTM) and canopy surface model (CSM), which were then used to estimate maize crop height.Furthermore, two strategies were explored — (1) grid expansion, and (2) using ground reference derived during sparse canopy aspotential tools to mitigate issues during dense canopy cover for multispectral image-based estimations. Results Results showed that the crop height estimated from multispectral images during sparse canopy cover show comparable accuracy (r = 0.9, nRMSE = 32%) to LiDAR-based estimation (r = 0.81, nRMSE = 17%). The accuracy of multispectral-based estimation, however, significantly declined under dense canopy cover, where no meaningful correlation was observed (r = − 0.1, nRMSE = 61%). In contrast, LiDAR-based estimates show consistent level of accuracy under the same conditions (r = 0.78, nRMSE = 15%). When grid expansion strategy was implemented for multispectral-based estimation during the dense canopy cover, the nRMSE was observed to decrease up to 28%. Meanwhile, with the use of sparse canopy cover ground reference, the nRMSE decreased up to 17%, comparable to the LiDAR-based estimate. Conclusion Overall, these results demonstrate the potential of deriving crop heights cost-effectively using multispectral imaging from UAS. This approach offers a scalable and economically viable alternative to expensive LiDAR systems, enabling wider integration of crop height information into precision agriculture applications.