Related Experiment Video
Updated: Oct 27, 2025
05:34
Analysis of Arabidopsis thaliana Growth Behavior in Different Light Qualities
Published on: February 2, 2018
19.3K
A Two-Step Phenotypic Parameter Measurement Strategy for Overlapped Grapes under Different Light Conditions
Yubin Miao1, Leilei Huang1, Shu Zhang1
1School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai 200240, China.
Sensors (Basel, Switzerland)
|July 20, 2021
Summary
This study introduces an improved HED network and contour fitting method for accurate grape particle area measurement, overcoming challenges of complex backgrounds and overlapping fruits for better growth analysis.
Area of Science:
- Agricultural Science
- Computer Vision
- Plant Phenotyping
Background:
- Accurate measurement of grape phenotypic characteristics, like fruit particle projection area, is crucial for assessing growth status and physiological changes.
- Complex backgrounds and overlapping fruit particles in images hinder precise border recognition and area calculation.
- Existing methods struggle with the accurate detection and measurement of individual grape particles in dense clusters.
Purpose of the Study:
- To develop a robust two-step method for accurately measuring the areas of individual grape particles, even when overlapped.
- To enhance the accuracy and efficiency of grape border recognition and phenotypic parameter extraction.
- To provide a reliable tool for analyzing grape growth laws through precise phenotyping.
Main Methods:
- An improved HED network utilizing multi-scale edge detection with Dice coefficients and image pyramids for enhanced particle edge detection.
- A novel contour fitting approach combining iterative least squares ellipse fitting and region growth algorithms for area calculation.
- Comparative experiments evaluating the improved HED network against Canny, HED, and DeepEdge for edge detection performance.
Main Results:
- The improved HED network demonstrated superior clarity, accuracy, and efficiency in extracting fruit particle edges compared to existing methods.
- The proposed method successfully detected overlapping grape contours more completely, addressing a key limitation in previous approaches.
- The shape-fitting step achieved a highly accurate grape area estimation with an average error of only 1.5%.
Conclusions:
- The developed two-step method significantly improves the accuracy and completeness of grape particle edge detection and area measurement.
- This approach offers a convenient and effective means for extracting grape phenotype characteristics and understanding grape growth dynamics.
- The findings provide valuable tools for precision agriculture and horticultural research focused on grape cultivation.
