A non-destructive coconut fruit and seed traits extraction method based on Micro-CT and deeplabV3+ model

Area of Science:

• Agricultural Science
• Biotechnology
• Imaging Technology

Background:

• Coconut breeding is advancing with gene mapping and molecular tools.
• Accurate measurement of coconut phenotypic traits is crucial for genotype-phenotype correlation.
• Molecular marker-assisted breeding requires precise phenotyping for effective screening.

Purpose of the Study:

• To develop an automated, non-destructive Micro-Computed Tomography (Micro-CT) system for measuring coconut fruits and seeds.
• To acquire 3D models and detailed phenotyping traits of coconuts.
• To provide technical support for coconut breeding programs.

Main Methods:

• Development of a Micro-CT system for automated, non-destructive measurement.
• Utilizing a deeplabv3+ model with an Xception backbone for automatic segmentation of sectional images.
• Comparison of Micro-CT measurements against structural-light system and manual measurements.

Main Results:

• Micro-CT system achieved low mean absolute percentage errors for fruit volume (1.87%) and surface area (2.24%) compared to structural-light systems.
• High correlation coefficients (0.977 for volume, 0.964 for surface area) were observed between Micro-CT and structural-light measurements.
• Micro-CT demonstrated good accuracy for copra weight (8.85% error) and total biomass (25.19% error) compared to manual measurements.
• The system can precisely obtain up to 21 agronomic and 57 digital traits non-destructively.

Conclusions:

• The developed Micro-CT system offers a precise and non-destructive method for coconut phenotyping.
• This technology is a valuable tool for molecular marker-assisted breeding, enabling accurate genotype-phenotype correlation.
• The system's ability to capture numerous traits supports advanced research in coconut genetics and breeding.