LI-YOLOv8: A lightweight bridge multi-defect detection algorithm

To address the challenges of high computational complexity and low accuracy in detecting small objects in multi-defect detection algorithms for bridge surfaces, a lightweight multi-defect detection algorithm based on an improved YOLOv8 is proposed. This approach balances detection network accuracy with the number of parameters, making the model easier to deploy. By discarding the P5 feature layer, the issue of large model parameters is mitigated, and the network becomes more suitable for detecting minor defects. The LMConv convolution was embedded into certain Bottleneck structures in the C2f module of YOLOv8n, addressing the problem of large parameter sizes. Additionally, the Inner-IoU loss function is introduced, using a larger scale factor to control the auxiliary boundary size to accelerate convergence and improve detection accuracy. The experimental results show that LI-YOLOv8 performs well on the self-built bridge surface defect dataset, reducing the number of parameters by 40.2% and the computational cost by 14.8%. Compared to the original YOLOv8n algorithm, mAP50 increased by 0.5%, and the detection speed reached 294.12 frames per second (FPS), achieving real-time and accurate detection of bridge surface defects on edge devices.