Pavement damage detection based on attention feature fusion iAFFNet
Pavement damage detection technology is critical to the safety and reliability of autonomous driving systems. To address the challenge related to the difficulty in balancing rapid detection with accurate localization,taking YOLOv8s network as the base-line,an improved pavement damage detection algorithm iAFFNet combining attention feature fusion was proposed. Firstly,the effi-cient channel attention (ECA) mosssdule was integrated into the shallower layers output of the backbone network to enhance the net-work's ability to focus on key features and effectively capture contextual information. Secondly,the iterative attentional feature fusion (iAFF) module was introduced in the enhanced feature extraction stage to improve localization accuracy. Finally,in order to verify the effectiveness of the proposed model,experimental evaluations are carried out on public and self-made datasets. The macro-F1 scores are improved by 2.88％ and 1.30％ respectively,while the mAP is improved by 0.43％ and 0.54％ respectively. The model parameters are only 12.749×106,which improves the detection performance without increasing the running time. More-over,frame per second (FPS) of the model achieves 32.7,meeting the requirements for real-time detection.
万方数据
维普
