Improved pavement crack recognition algorithm of YOLOv8
Road surface cracks are one of the critical hidden dangers affecting the normal use of roads and traffic safety.To address the issues of high cost,low efficiency,and insufficient accuracy in the current algorithms for pavement crack identification,an improved pavement crack identification algorithm based on the YOLOv8 framework is proposed.Firstly,a small object layer and an additional detection segmentation head are introduced to enhance the detection and fusion capabilities of local fine feature information.Secondly,by borrowing the context correlation capabilities of the Transformer in processing sequential data,the PET module is integrated to obtain a global self-attention mechanism,further optimizing the identification performance for fine and long cracks.Additionally,SPPF reuse is introduced to enhance the representation of feature information,improving the recognition and localization capabilities of target objects.The results show that the improved model significantly enhances pavement crack identification,with an mAP50 of 73.1%,representing an 8.3%improvement compared to the original model.Meanwhile,a comparative analysis with four other algorithms,SSD,Mask R-CNN,YOLOv5,and YOLOv6,demonstrates that the improved algorithm achieves higher recognition accuracy and environmental adaptability while balancing temporal-spatial resource consumption and accuracy.
deep learningroad surface crack detectionYOLOv8 modeldetection headPET module
万方数据
维普
