Bridge Deflection Monitoring Method Based on Yolo v8-DCC

Using computer vision technology to measure bridge deflection offers advantages of non-contact, high precision, and real-time monitoring. However, existing methods exhibit insufficient accuracy and poor stability when monitoring dynamic changes in deflection. To address these limitations, this work proposes a monitoring method based on Yolo v8 with Deformable Convolutional Networks (DCN) and Convolutional Block Attention Module (CBAM) mechanism (Yolo v8-DCC). This method utilizes image segmentation techniques, employing circular markers as monitoring targets to capture detailed bridge deformations, improving the accuracy and stability of deflection measurement. In this work, the traditional convolutional modules in the C2f module of the original Yolo v8 algorithm was replaced with a DCN module, resulting in the C2f-DCN module. By adaptively adjusting the shape and position of the convolutional kernel, the algorithm enhances feature extraction across varying scales and complex backgrounds, preventing the loss of critical edge feature information. Furthermore, this work added CBAM to the Yolo v8 algorithm to refine the attention distribution of the model, focusing on the basic regions in the segmentation task and improving the accuracy of object detection. Experimental results demonstrate that Yolo v8-DCC achieves a mean Average precision (mAP) of 99.2% at IoU threshold 0.95, representing a 2.5% improvement over the original Yolo v8 model, with overall accuracy enhanced by 5%. Compared to other segmentation algorithms, Yolo v8-DCC significantly improves both precision and stability, facilitating effective and precise monitoring of bridge deflection.