Re-Enhancing Multi-Target Classification Techniques for CODEBRIM Dataset using Deep Learning Algorithm

The existence of concrete defects such as corrosion stains, cracks, efflorescence, exposed bars, and spallation has greatly compromised the safety and convenience of bridge users. These calamities can lead to deaths and have an impact on both communities and the economy. Effective maintenance management relies on the receipt of timely and accurate information. Technological progress facilitates the instantaneous handling of up-to-the-minute data. The objective of this work is to enhance the performance of YOLO models for current applications. The CODEBRIM custom dataset is used, which consists of $\mathbf{7, 2 6 1}$ photographs illustrating concrete flaws. The results demonstrated that the YOLOv9m version attained the highest levels of accuracy and precision, reaching 74.99% and $89.41 \%$ correspondingly. Both YOLOv9 and YOLOv10 are outperformed in terms of accuracy, but they excel in training time, with YOLOv9 producing a more convergent result.