基于注意力机制和多尺度特征融合的道路裂缝检测

Road crack detection based on attention mechanism andmulti-scale feature fusion

针对复杂背景下细小裂缝难以检测和裂缝检测出现断裂的问题,提出一种基于注意力机制和多尺度特征融合的道路裂缝检测方法EAFNet。在编码阶段设计边缘细化模块,更好提取裂缝细节特征;在网络中间部分设计基于注意力机制的多尺度特征融合模块对裂缝进行准确定位;在解码器部分设计融合优化模块,更好提取裂缝特征和定位裂缝位置。在公开数据集CRACK500训练集上进行训练并在两个道路裂缝数据集上进行测试,与现有的部分检测方法相比,该算法在分割精度和泛化性上都有提升,该算法对于细小裂缝的分割更为精细且有效解决了裂缝检测的断裂问题。

Aiming at the problems that small cracks are difficult to detect and crack detection breaks in complex background,a road crack detection method EAFNet based on attention mechanism and multi-scale feature fusion was proposed.In the coding stage,an edge refinement module was designed to better extract the detailed features of cracks.A multi-scale feature fusion module based on attention mechanism was designed in the middle part of the network to accurately locate cracks.A fusion optimization module was designed in the decoder part to better extract crack features and locate crack locations.The algorithm was trained on the public dataset CRACK500 training set and tested on two road crack datasets.Compared with the existing partial detection methods,the algorithm improves segmentation accuracy and generalization,indicating that the algorithm is more precise for the segmentation of small cracks and effectively solves the fracture problem of crack detection.