Adherent Peanut Image Segmentation Based on Multi-Modal Fusion.

Area of Science:

• Computer Vision
• Agricultural Engineering
• Image Processing

Background:

• Segmenting adherent peanut pods is challenging due to their non-convex shapes, complex textures, and varied structures.
• Accurate segmentation is crucial for automated agricultural processes and quality assessment.

Purpose of the Study:

• To develop a multimodal fusion algorithm for accurate 2D segmentation of adherent peanut images.
• To overcome limitations of traditional segmentation methods for irregularly shaped objects.

Main Methods:

• Utilized a line structured light imaging system to capture 3D point clouds of peanuts.
• Employed surface fitting, minimum curvature-based seed point selection, and KD-Tree algorithm for point cloud segmentation.
• Extracted 2D contours using a rolling method and optimized segmentation with multiscale feature matching and opening operations.

Main Results:

• The proposed algorithm successfully segmented individual peanut pods from adherent images.
• Achieved a high segmentation accuracy of 96.8% for peanut pod images.
• Demonstrated improved performance compared to existing segmentation techniques.

Conclusions:

• The multimodal fusion approach effectively addresses the challenges of segmenting adherent peanut images.
• This method provides a robust solution for accurate peanut pod segmentation in agricultural applications.