Scholarlink

<p>Remote sensing technologies, particularly unmanned aerial vehicles (UAVs) equipped with multispectral sensors, have transformed high-throughput plant phenotyping by enabling the efficient and repeatable monitoring of plant traits across large experimental plots. These tools are vital for breeders developing stress-resilient cultivars across diverse environments. However, accurately aligning UAV-derived orthophotos across time remains a&#xa0;key challenge, especially in field conditions where ground control point (GCP) placement is infeasible due to dynamic crop growth, field operations, or resource constraints. To address this, we present a&#xa0;cost-effective solution for registering UAV-derived orthophotos acquired at different time-points in breeding trials, without relying on high-precision global navigation satellite system (GNSS) or extensive GCP-based infrastructure. Our method leverages the Segment Anything Model for initial crop plot segmentation, followed by a&#xa0;custom feature extractor, descriptor, and matcher designed to exploit the stable geometric layout of crop fields. This centroid-based registration approach achieves RMSE between 4.65 and 12.30 cm, sufficient for plot-level analysis in many agronomic applications. Unlike traditional workflows that demand sub-centimeter accuracy through RTK-GNSS, our approach prioritizes operational ease and robustness under practical constraints. Furthermore, we validate the method across three crops, including wheat, rice, and sunflower grown under checkerboard, grid, and strip plot layouts. The NDVI values obtained using our registered orthophotos show strong agreement with those measured by the handheld GreenSeeker device, as reflected by high correlation and low RMSE, confirming that our registration method enables reliable plot-level vegetation monitoring. This makes our pipeline particularly well-suited for resource-limited research settings where maintaining precise GCPs is impractical.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Advancing Spatiotemporal Orthophoto Registration in UAV-based Crop Breeding Experiments Leveraging Field Geometric Features

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • Muhammad Salman Akhtar,
  • Zuhair Zafar,
  • Zahid Mahmood,
  • Haris Khurshid,
  • M. R. Naseem,
  • Karsten Berns,
  • Muhammad Moazam Fraz

摘要

Remote sensing technologies, particularly unmanned aerial vehicles (UAVs) equipped with multispectral sensors, have transformed high-throughput plant phenotyping by enabling the efficient and repeatable monitoring of plant traits across large experimental plots. These tools are vital for breeders developing stress-resilient cultivars across diverse environments. However, accurately aligning UAV-derived orthophotos across time remains a key challenge, especially in field conditions where ground control point (GCP) placement is infeasible due to dynamic crop growth, field operations, or resource constraints. To address this, we present a cost-effective solution for registering UAV-derived orthophotos acquired at different time-points in breeding trials, without relying on high-precision global navigation satellite system (GNSS) or extensive GCP-based infrastructure. Our method leverages the Segment Anything Model for initial crop plot segmentation, followed by a custom feature extractor, descriptor, and matcher designed to exploit the stable geometric layout of crop fields. This centroid-based registration approach achieves RMSE between 4.65 and 12.30 cm, sufficient for plot-level analysis in many agronomic applications. Unlike traditional workflows that demand sub-centimeter accuracy through RTK-GNSS, our approach prioritizes operational ease and robustness under practical constraints. Furthermore, we validate the method across three crops, including wheat, rice, and sunflower grown under checkerboard, grid, and strip plot layouts. The NDVI values obtained using our registered orthophotos show strong agreement with those measured by the handheld GreenSeeker device, as reflected by high correlation and low RMSE, confirming that our registration method enables reliable plot-level vegetation monitoring. This makes our pipeline particularly well-suited for resource-limited research settings where maintaining precise GCPs is impractical.