Fragment hole image segmentation algorithm of target plate using PIDNet with multi-scale fusion dual attention

Area of Science:

• Engineering
• Materials Science
• Computer Vision

Background:

• Ineffective segmentation of perforations in static explosion damage experiments leads to loss of edge information and low accuracy.
• Accurate segmentation is crucial for analyzing fragment dispersion and destructive effects.

Purpose of the Study:

• To develop an improved fragment hole image segmentation algorithm for target plates.
• To enhance the accuracy and preserve edge characteristic information in explosion damage analysis.

Main Methods:

• Introduced a fragment hole image segmentation algorithm utilizing PIDNet with multi-scale fusion dual attention.
• Employed a PID triple-branch network architecture with a dual-attention module to enhance hole edge features.
• Integrated a multi-scale feature fusion (MSFF-Bag) module to improve hole contour segmentation accuracy.

Main Results:

• The proposed algorithm demonstrated superior performance compared to state-of-the-art methods.
• Achieved higher mean intersection and concurrency ratio and pixel accuracy.
• Successfully preserved critical edge characteristic information for detailed analysis.

Conclusions:

• The developed algorithm provides a reliable method for segmenting fragment holes in target plates.
• Offers a guarantee for analyzing fragment dispersion characteristics and destructive effects.
• Holds significant military theoretical value and practical application importance.