A Structural Re-Parameterization Multiple Object Tracking Algorithm for Underwater Unmanned Autonomous Systems

Complex underwater environments can seriously affect the stability of the imaging equipment and the quality of the acquired images of underwater unmanned autonomous systems. Considering the impact of the above factors on the effectiveness of underwater multiple object tracking, this paper proposes a structural re-parameterization multiple object tracking algorithm that is applicable to underwater unmanned autonomous systems. Firstly, a multi-scale feature extraction convolutional structure based on structural reparameterisation (DBB-RFAConv) is in-troduced into the neck of YOLOv8 to enhance the multi-scale feature extraction capability of the network and improve the accuracy of underwater target detection. Secondly, Inner-PIoU is used as the loss function of YOLOv8 to optimize the internal overlap between the prediction boxes and the ground truth boxes to strengthen the robustness, convergence speed and tracking accuracy of the model in underwater detection and tracking tasks. Finally, a novel multiple object tracking algorithm for underwater unmanned autonomous systems is proposed, which incorporates the improved YOLOv8 algorithm with the BoT-SORT algorithm. The experimental results verify the excellent performance of the proposed algorithm in complex underwater environments.