FLDO-LKNet: An Efficient Method for Segmenting Stone Cells in Rubber Tree Bark with Robustness to Staining...

Area of Science:

• Plant anatomy
• Agricultural technology
• Biomedical image analysis

Background:

• Stone cells are crucial structural components in rubber tree bark, influencing key agricultural traits.
• Existing segmentation methods lack robustness against staining variations, diverse cell morphology, and scale changes in histological images.

Purpose of the Study:

• To develop a high-precision, automated semantic segmentation network for rubber tree stone cells.
• To address limitations in current methods for segmenting stone cells in challenging histological images.

Main Methods:

• Proposed FLDO-LKNet, an automated semantic segmentation network for stone-cell delineation.
• Introduced LDFE module for channel-level feature recalibration to handle staining variations.
• Designed KCFA attention mechanism to capture complex morphology and scale variations.
• Developed FLDO optimization algorithm with dynamic learning rate adjustment for robustness against debris interference.
• Created a dataset of 1084 stone-cell images for training and evaluation.

Main Results:

• FLDO-LKNet achieved 75.18% mIoU, 98.5% accuracy, and 82.21% sensitivity.
• The method demonstrated high-precision pixel-level segmentation of stone cells.
• The developed modules effectively mitigated staining inconsistencies, morphological diversity, and debris interference.

Conclusions:

• FLDO-LKNet offers a robust solution for precise stone-cell segmentation in rubber tree histology.
• This advancement facilitates further research into stone cell development, genetic functions, and agricultural applications.