On March 3, researchers Chen Yaohui and Li Weifu from Huazhong Agricultural University (HZAU) published a breakthrough study in Plant Phenomics, introducing "InspectGaussian". This novel coarse-to-fine reconstruction framework establishes a critical foundation for robotic inspection and precision management in complex orchard environments.

The research addresses long-standing bottlenecks in large-scale 3D modeling, such as variable illumination and computational constraints. A key innovation is the data acquisition strategy for inspection robots. By replacing traditional plant-centric circling with an efficient inter-row peripheral shooting approach, the team solved multi-view capture problems in densely planted areas.

Technically, the framework integrates RGB-D-based SLAM with a customized dense mapping module. This integration ensures accurate pose estimation and precise point cloud generation, effectively mitigating interference caused by shifting light and differing plant sizes.

Furthermore, the team applied a "divide-and-conquer" strategy, combining 3D Gaussian Splatting (3DGS) with depth regularization and region-aware refinement to achieve high-fidelity reconstruction of individual plants.

Experimental results in citrus orchards confirm the system's superior performance, reaching an impressive 31.226 PSNR and 7 mm point cloud accuracy.

By capturing fine structural details while maintaining scalability, InspectGaussian provides a practical, high-throughput solution for in-field plant phenotyping. This advancement greatly strengthens the capabilities of next-generation intelligent orchard management systems.

The orchard reconstruction framework. [Photo/news.hzau.edu.cn]