Future work will focus on expanding the benchmark dataset to include more crack scenarios and background images. Additionally, we plan to investigate the use of our benchmark for evaluating the performance of other materials science-related algorithms, such as those for detecting defects and corrosion.
Recently, several crack detection algorithms have been proposed, including those based on image processing, machine learning, and deep learning techniques. While these algorithms have shown promising results, their performance is often evaluated using different datasets and metrics, making it difficult to compare their effectiveness.
The results of the verification study are presented in Tables 1-3, which show the performance of each algorithm under different crack conditions.
In this paper, we presented a novel superposition benchmark for verifying crack detection algorithms. Our benchmark provides a standardized framework for evaluating the performance of crack detection algorithms, allowing for a thorough assessment of their effectiveness. We demonstrated the effectiveness of our benchmark by verifying several state-of-the-art crack detection algorithms and analyzing their performance under different conditions. The results show that our benchmark is effective in evaluating the performance of crack detection algorithms and can be used to identify the most effective algorithms for specific applications.
| Algorithm | Precision | Recall | F1-score | MAP | | --- | --- | --- | --- | --- | | Image processing-based | 0.8 | 0.7 | 0.75 | 0.85 | | Machine learning-based | 0.9 | 0.8 | 0.85 | 0.9 | | Deep learning-based | 0.95 | 0.9 | 0.925 | 0.95 |