CCMIM: Optimizing concrete defect detection through state-space modeling and dynamic feature fusion.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Concrete defect detection is vital for structural safety and durability.
• Convolutional Neural Networks (CNNs) struggle with multi-scale features and global context.
• Transformer models are computationally intensive, limiting real-time applications.

Purpose of the Study:

• To propose a novel end-to-end framework, Concrete Crack Mamba-in-Mamba (CCMIM), for efficient and accurate concrete defect detection.
• To enhance the model's ability to capture long-range dependencies and global context.
• To improve robustness, adaptability, and reduce computational cost for real-time detection.

Main Methods:

• Introduced the Mamba-In-Mamba (MiM) module for hierarchical data flow and long-range dependency capture.
• Proposed the Dynamic Dual Fusion (DDF) module for robust multi-scale feature fusion.
• Developed the Sparse Pyramid Transformer (SPT) module to reduce computation and maintain spatial information while improving inference speed.

Main Results:

• The CCMIM model demonstrated superior performance compared to traditional, YOLO-, and Transformer-based methods in concrete crack detection.
• Achieved high accuracy rates on multiple datasets: 89.2% on RDD2022, 85.2% on SDNET2018, and 79.3% on CCCD.
• Obtained excellent mAP50 scores: 88.1% on RDD2022, 87.8% on SDNET2018, and 79.2% on CCCD.

Conclusions:

• The CCMIM framework offers an effective solution for concrete defect detection, outperforming existing methods.
• The proposed MiM, DDF, and SPT modules contribute to improved accuracy, robustness, and efficiency.
• CCMIM provides a promising approach for real-time structural health monitoring and defect analysis.