基于改进YOLOv8的道路病害视觉识别算法

Road damage visual recognition algorithm based on improved YOLOv8

道路病害检测对于确保道路的安全性和可持续性至关重要,对城市和社会的发展具有积极作用。为提高目前道路病害检测模型的性能,文中提出一种基于改进YOLOv8的道路病害检测算法。设计一种新型高效的特征融合模块(DWS),提高模型获取特征信息和全局上下文信息的能力;提出将ECABlock、LeakyReLU激活函数与卷积相结合的新模块ELConv来提高深层网络对目标的定位能力;另外,使用Dynamic Head检测头替换原始YOLOv8的头部,结合尺度、空间和任务三种注意力机制提升模型头部表征能力;最后,采用WIoU损失函数代替原损失函数来改善边界框精确度和匹配度。相比基线模型,改进模型在road damage detection数据集和RDD2022_Japan数据集上都得到了有效的验证,表明改进模型满足当下道路病害检测的需求,展示了高灵活性、准确性和效率。

Road damage detection is crucial to ensure the safety and sustainability of roads,and plays a positive role in the development of cities and society.A road damage detection algorithm based on improved YOLOv8 is proposed to improve the performance of the current road damage detection model.A new and efficient feature fusion module DWS is designed to enhance the model ability of obtaining feature information and global context information.A new module ELConv,which combines activation functions ECABlock and LeakyReLU with convolution,is proposed to improve the deep network's ability to locate objects.In addition,the detection head Dynamic Head is used to replace the head of the original YOLOv8.The three attention mechanisms of scale,space and task are combined to improve the representation ability of the model head.The loss function WIoU is used to replace the original loss function to improve the accuracy and match of the bounding box.In comparison with the baseline model,the improved model is verified on both the road damage detection dataset and the RDD2022_Japan dataset effectively.It shows that the improved model can meet the current needs of road damage detection,demonstrating high flexibility,accuracy and efficiency.