Growth parameter acquisition and geometric point cloud completion of lettuce

Area of Science:

• Agricultural Engineering
• Computer Vision
• Plant Science

Background:

• Controlled Environment Agriculture (CEA), including plant factories, is crucial for global food security.
• Accurate measurement of plant growth parameters is essential for process control and yield prediction in plant factories.
• Existing methods for growth parameter extraction can be destructive or lack precision, especially with occlusions.

Purpose of the Study:

• To develop a fast, non-destructive framework for extracting plant growth parameters in plant factories.
• To propose a geometric point cloud completion method to address occlusion issues in 3D plant scans.
• To validate the correlation between the proposed method's output and actual plant growth metrics.

Main Methods:

• Utilized a Time-of-Flight (ToF) camera (Microsoft Kinect V2) to capture top-view 3D point clouds of lettuce.
• Developed a geometric method to process and complete incomplete point clouds based on lettuce growth characteristics.
• Separated lettuce point clouds and analyzed the linear correlation between processed point cloud data and measured plant height, leaf area, and fresh weight.

Main Results:

• The geometric point cloud completion method demonstrated a high linear correlation with actual plant height (R² = 0.961).
• Strong correlations were also found for leaf area (R² = 0.964) and fresh weight (R² = 0.911).
• The treated point cloud data showed significant improvement over untreated point cloud data in predicting growth parameters.

Conclusions:

• The proposed point cloud completion framework provides a fast and non-destructive approach for plant growth parameter estimation.
• This method effectively handles occlusions from a single 3D view, offering a practical solution for plant factories.
• The findings suggest a high potential for this technique in optimizing CEA operations and improving yield estimation.