Petal segmentation in CT images based on divide-and-conquer strategy.

Area of Science:

• * Botanical imaging and analysis.
• * Computer vision and machine learning applications in morphology.

Background:

• * Manual segmentation of flower petals in computed tomography (CT) images is laborious.
• * Existing instance segmentation methods struggle with the unique petal shapes in CT data.

Purpose of the Study:

• * To develop an automated petal segmentation method for 3D Camellia japonica flower reconstruction.
• * To improve the accuracy and efficiency of petal segmentation in CT images.

Main Methods:

• * Proposed a petal segmentation approach using computer vision techniques on CT slice images.
• * Implemented a cropping strategy, extracting 2D long rectangles from each slice to simplify segmentation.
• * Applied instance segmentation methods to cropped images and integrated results for 3D reconstruction.

Main Results:

• * The proposed cropping method significantly enhanced petal segmentation accuracy on 2D slice images compared to non-cropped methods.
• * Successful generation and visualization of 3D segmentation volume data for Camellia japonica flowers.
• * Training dataset augmentation via cropping enabled the use of advanced segmentation models.

Conclusions:

• * The novel cropping-based computer vision method effectively addresses challenges in CT image petal segmentation.
• * This technique facilitates accurate 3D reconstruction and visualization of flower structures.
• * Offers a more efficient and accurate alternative to manual segmentation for botanical CT imaging.