3D Quantitative Modeling for Stone Fruit Quality Assessment by LF-NMRI

Area of Science:

• Agricultural Engineering
• Biomedical Engineering
• Computer Vision

Background:

• Core Volume Ratio (CVR) is crucial for assessing edible fruit fractions.
• Traditional CVR measurement involves destructive sampling.
• Low-field Nuclear Magnetic Resonance Imaging (LF-NMRI) offers non-destructive analysis but faces spatial resolution limitations.

Purpose of the Study:

• To develop a high-precision 3D modeling method for non-destructive internal quality analysis of stone fruits.
• To overcome the spatial quantification limitations of LF-NMRI in fruit analysis.
• To accurately determine the Core Volume Ratio (CVR) using a novel 3D reconstruction approach.

Main Methods:

• Acquisition of tomographic LF-NMRI sequences along three orthogonal axes.
• Segmentation of fruit pulp and core regions using a SwinUNet deep learning model.
• Registration of point clouds from three views using a genetic algorithm and fusion into a unified 3D model via Poisson surface reconstruction.

Main Results:

• Achieved a highly accurate CVR estimation with a mean absolute error of 0.13% compared to manual measurements.
• The three-view reconstruction strategy resulted in a volumetric error of only 0.73%.
• Significantly outperformed single-view (4.57%) and dual-view (3.73%) reconstruction approaches.

Conclusions:

• The proposed method provides a robust and accurate non-destructive solution for 3D internal fruit quality analysis.
• This technology enables precise quantification of core and entire fruit volumes.
• The SwinUNet and multi-view reconstruction approach enhance the capabilities of LF-NMRI for agricultural applications.