Automated Phenotypic Trait Extraction for Rice Plant Using Terrestrial Laser Scanning Data

Kexiao Wang, Xiaojun Pu, Bo Li2024 · DOI

TL;DR

This paper introduces an automated tool for extracting key phenotypic traits of rice plants using terrestrial laser scanning (TLS) data. By developing a novel three-layer architecture and integrating advanced point cloud processing algorithms, the tool efficiently quantifies parameters like plant height, crown diameter, and tiller number, offering a significant advancement over traditional manual measurement methods.

What Problem Does This Paper Address?

Accurately and efficiently measuring plant phenotypes is crucial for crop breeding and understanding plant growth. Traditional methods, often involving manual measurements, are time-consuming and labor-intensive, creating a bottleneck in research. While remote sensing technologies like LiDAR have shown promise, extracting detailed phenotypic information from complex plant structures, such as rice with its flat leaves and numerous tillers, remains challenging. Existing research often focuses on population-level characteristics or simpler plant structures, leaving a gap in automated, high-throughput extraction of individual rice plant traits.

Methods

The study proposes a multi-step approach for automated phenotypic trait extraction from TLS point cloud data. The core methodology involves:

  • Point Cloud Preprocessing: This includes data collection, registration, stitching, and filtering (e.g., statistical outlier removal) to clean and prepare the raw point cloud data.
  • Phenotypic Parameter Calculation: A suite of algorithms is employed for extracting seven key traits:
  • Crown Diameter and Perimeter of Stem: These are calculated using a "neighborhood search algorithm" based on Kd-trees to find distances from the centroid of the stem cross-section to the farthest points of the canopy projection.
  • Plant Height: Determined as the vertical distance from the highest point of the plant to a reference point.
  • Surface Area: Estimated using a 3D surface reconstruction approach with the Alpha-shape algorithm.
  • Volume: Calculated using the minimum convex hull that encloses the plant.
  • Projected Leaf Area (PLA): Determined by calculating the projection area of the plant and comparing it to the minimum bounding rectangle of the canopy's horizontal projection.
  • Tiller Number: A novel approach is presented, utilizing Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) on segmented stem point clouds to identify and count tiller branches.
  • Automated Extraction Tool Development: A user-friendly tool was developed using the PyQt5 framework for the graphical user interface (GUI) and the Open3D library for 3D point cloud processing. This tool integrates the developed algorithms into a cohesive system.
  • Accuracy Evaluation: The accuracy of the extracted parameters was assessed by comparing them with manually measured values using linear regression, calculating R², Root Mean Square Error (RMSE), and Relative Root Mean Square Error (RRMSE).

Datasets

The study utilized point cloud data of rice plants collected indoors using a FARO Focus S70 terrestrial laser scanner. Specific details about the scanner's resolution and measurement error are provided. For validation, manual measurements of crown diameter, perimeter of stem, plant height, and tiller number were performed using measuring tapes and manual counting. The paper notes that the accuracy verification was conducted on a "relatively small sample size" of six rice plant point clouds, which is identified as a limitation.

Key Insights and Ideas

  • Automated Tiller Number Extraction: The development of an HDBSCAN-based method for automatically counting tiller numbers is a significant contribution, addressing a particularly challenging trait due to rice plant complexity.
  • Integrated Phenotyping Tool: The creation of a user-friendly, three-layer architecture tool that combines data interaction, basic processing, and functional modules offers a comprehensive solution for rice plant phenotyping.
  • Efficiency Gains: The tool demonstrates the potential for rapid and efficient acquisition of phenotypic data, which is critical for accelerating crop breeding programs.
  • Robustness of Algorithms: The study highlights the effectiveness of algorithms like Alpha-shape for surface reconstruction and HDBSCAN for density-based clustering in handling complex 3D plant structures.

Key Results

The study reports promising results for the automated extraction of rice plant phenotypic parameters:

  • High Correlation: R² values for the linear correlation between measured and extracted values were generally high, with plant height reaching 0.97, crown diameter and tiller number above 0.80, and perimeter of stem at 0.66.
  • Centimeter-Level Accuracy: RMSE for crown diameter, perimeter of stem, and plant height were within the centimeter range.
  • Low Tiller Number RMSE: The RMSE for tiller number was notably low at 1.63.
  • Low RRMSE for Key Traits: RRMSE values for crown diameter, plant height, and tiller number were within 10%, indicating minimal difference from measured values.
  • Perimeter of Stem RRMSE: The RRMSE for the perimeter of stem was 18.29%, which the authors attribute to potential differences in measurement positions.
  • Overall Reliability: The tool's extraction algorithm was found to exhibit good robustness and accuracy, providing a reliable method for obtaining phenotypic characteristics.

Limitations and Future Work

The authors acknowledge several limitations and outline future research directions:

  • Sample Size: The accuracy verification was performed on a limited number of rice plant sample point clouds. The authors plan to systematically collect more data to improve feature extraction and comparison verification.
  • Consistency in Measurements: The RRMSE for stem circumference was slightly higher due to difficulties in ensuring consistent stem slice selection and manual measurement positions.
  • Point Cloud Completeness: For rice plants with many tillers, point clouds can be incomplete due to obstructions, which can affect tiller number extraction.
  • Algorithm Improvement: Future work will focus on improving the accuracy of the extraction algorithms.
  • Broader Applicability: While developed for rice, the tool's indicator extraction functions are expected to have wide applicability to other crop types.

Summary updated

BibTeX
@misc{wang2024automated,
  author = {Kexiao Wang and Xiaojun Pu and Bo Li},
  title = {{Automated Phenotypic Trait Extraction for Rice Plant Using Terrestrial Laser Scanning Data}},
  year = {2024},
  doi = {10.3390/s24134322},
}
Figure