A high-precision segmentation method for rubber tree stone cells

Area of Science:

• Plant biology
• Biotechnology
• Computer vision

Background:

• Stone cells in rubber tree bark are crucial for traits like hardness and latex yield.
• Current methods for stone cell analysis are slow and inaccurate, hindering research.
• Understanding stone cell distribution is vital for rubber tree cultivation and genetic studies.

Purpose of the Study:

• To develop an automated, accurate method for segmenting and quantifying stone cells in rubber tree bark.
• To improve the analysis of stone cell distribution and its impact on rubber tree traits.
• To overcome limitations of traditional segmentation networks in handling complex bark image features.

Main Methods:

• Proposed CGWO-LWNet, an automatic segmentation network using image recognition.
• Introduced a low-rank KAN module for enhanced feature fusion and edge segmentation.
• Designed a wavelet attention mechanism (Wave-SC) for capturing stone cell distribution patterns.
• Utilized a novel gray wolf constrained optimization algorithm (CGWO) for stable network training.
• Created a dataset of 1084 rubber tree stone cell images for training and validation.

Main Results:

• CGWO-LWNet achieved 69.1% MIoU, 81.7% DSC, and 80.4% recall on the test dataset.
• Demonstrated superior accuracy with 97.8% for rubber tree bark stone cell segmentation compared to other algorithms.
• The network effectively handles complex edges and regional distribution patterns of stone cells.

Conclusions:

• CGWO-LWNet provides a practical and robust tool for high-precision stone cell segmentation in rubber tree bark.
• Enables large-scale, accurate trait analysis related to latex yield, bark integrity, and stress resilience.
• Facilitates further genetic studies on stone cell development and its influence on rubber tree productivity.