Low-Contrast Coating Surface Microcrack Detection Using an Improved U-Net Network Based on Probability Map Fusion.

Area of Science:

• Materials Science
• Computer Vision
• Image Processing

Background:

• Microcrack detection in coatings is challenging due to low contrast, complex backgrounds, and discontinuous crack patterns.
• Existing methods struggle with precise identification and characterization of fine or fragmented cracks.

Purpose of the Study:

• To develop an advanced method for accurate microcrack detection on coating surfaces.
• To enhance the robustness and sensitivity of crack detection algorithms.

Main Methods:

• A novel approach combining circular neighborhood features with an improved U-net (ImF-Att-DO-U-net) architecture.
• Preprocessing involves median filtering, difference operations, and Gaussian smoothing for contrast enhancement.
• Multi-scale, multi-directional circular scanning filters generate crack probability maps, used as dual-channel input for the U-net.
• The U-net encoder uses CBAMs, while the decoder incorporates DO-Conv and Leaky ReLU for improved feature extraction.
• A hybrid loss function (Binary Cross-Entropy and Dice loss) addresses class imbalance.

Main Results:

• The proposed method achieved a Dice coefficient of 0.884, SSIM of 0.893, and accuracy of 0.911.
• Successfully extracted 98% of cracks ≥10 μm, with a minimum detectable crack size of 7 μm.
• Demonstrated superior performance compared to existing models like DO-U-net and excellent robustness against noise and blur.

Conclusions:

• The ImF-Att-DO-U-net method significantly improves microcrack detection accuracy and sensitivity on coating surfaces.
• The approach offers robust performance and adaptability to various environmental conditions, including noise and blur.
• This technique provides a reliable tool for quality control and material integrity assessment in coating applications.