ENHANCED YOLO8 FOR DECTECTING MULTIPLE DEFECTS ON BRIDGE

In this study, an enhanced version of the YOLOv8 model is proposed by integrating a Convolutional Block Attention Module (CBAM) and a Wise-IoU loss function. This is a new model called YOLOv8-CBAM-Wise-IoU. The CBAM mechanism enhances feature representation by using both channel and spatial attention. This allows the model to focus on important parts of the image, and ignore background noise that is not important. This is especially important for detecting small cracks, subtle corrosion or overlapped defects that the standard detection models might miss. The Wise-IoU loss function improves the bounding box regression by dynamically adjusting the loss weight according to the object characteristics. This improves the accuracy of localisation and reduces the number of false positives.The proposed model can detect seven types of defects on bridge surfaces simultaneously. These include cracking, spalling, corrosion, delamination, surface scaling, efflorescence, and potholes. A large set of labelled pictures of bridges was collected and augmented using techniques such as rotation, scaling and changing brightness to increase the generalisability of the model. We did a lot of work to improve the training, such as tuning hyperparameters and ablation studies, to see how each part (YOLOv8 backbone, CBAM and Wise-IoU) affected the overall performance.The tests demonstrate that YOLOv8-CBAM-Wise-IoU outperforms standard YOLO models and other common methods. The model could detect 97.9% of the defects, 76% of the time it should have, 58% of the time it should have, and 55.4% of the time it should have. It could also detect defects in real-world conditions such as varying lighting, complex backgrounds and uneven surface textures. The system also had a faster detection speed that was good for real-time use, which means it could be used for inspection systems using drones or cameras.