A texture enhanced attention model for defect detection in thermal protection materials

Area of Science:

• Materials Science
• Aerospace Engineering
• Computer Vision

Background:

• Internal defects in thermal protection materials are critical for aerospace safety.
• Existing defect detection models face challenges with defect-background similarity, tiny defects, and multi-scale characteristics.
• A lack of real-world defect datasets hinders model development.

Purpose of the Study:

• To develop an accurate and efficient defect detection model for thermal protection materials.
• To address limitations in detecting small, multi-scale, and low-contrast defects.
• To create a comprehensive dataset for training and evaluating defect detection models.

Main Methods:

• Construction of a thermal protection material digital radiographic (DR) image dataset (TPMDR-dataset).
• Proposal of a texture-enhanced attention defect detection (TADD) model.
• Integration of a texture enhancement module, non-local dual attention, and path aggregation network.

Main Results:

• The TADD model achieved higher mean Average Precision (mAP) on the TPMDR-dataset and public datasets.
• The model demonstrated real-time detection capabilities at 25 frames per second.
• Performance exceeded the baseline model by 11.05%.

Conclusions:

• The TADD model offers a significant advancement in detecting internal defects in thermal protection materials.
• The developed dataset and model contribute to improved structural integrity and safety in aerospace applications.
• The TADD model effectively handles challenges posed by defect size, scale, and feature visibility.