基于聚类分割的裂缝检测方法

Crack detection method based on clustering segmentation

针对道路表面颗粒噪声干扰严重,传统聚类算法提取的裂缝轮廓不完整且连续性较差等问题。本文提出一种超像素及快速模糊C-均值聚类的路面裂缝提取方法,首先利用直方图均衡化提高正常路面和路面裂缝的对比度;然后,采用Gabor纹理特征结合简单线性迭代聚类方法生成超像素;最后,根据直方图信息计算隶属度矩阵,将裂缝超像素块聚合成完整的裂缝区域。依靠融合特征作为相似性度量的超像素分割方法,可提升超像素初始分割效果,获得更准确的超像素边缘。快速模糊C-均值聚类算法通过将直方图信息引入到目标函数,减少聚类分割的计算量,提升了计算效率。为证明所提算法的准确性,在公共数据集Crack Forest Dataset、DeepCrack上,与其他3种方法进行对比,实验结果表明本文算法能够有效地检测和提取路面裂缝,可为道路灾害检测提供有效信息。

Aiming at the serious interference of particle noise on road surface,the crack contour extracted by traditional clustering algorithm is incomplete and has poor continuity.This paper proposes a superpixel and fast fuzzy C-means clustering method for pavement crack extraction.Firstly,histogram equalization is used to improve the contrast between normal pavement and pavement cracks.Then,Gabor texture features are combined with simple linear iterative clustering method to generate superpixels.Finally,the fast fuzzy C-means clustering algorithm is used to calculate the membership matrix according to the histogram information,and the crack superpixel blocks are aggregated into a complete crack area.The superpixel segmentation method based on fusion feature as similarity measure can improve the initial segmentation effect and obtain more accurate superpixel edge.By introducing the histogram information into the objective function,the fast fuzzy C-means clustering algorithm reduces the computational complexity of clustering segmentation and improves the computational efficiency.In order to prove the accuracy of the proposed algorithm,it is compared with the other three methods on the public datasets Crack Forest Dataset and DeepCrack.The experimental results show that the proposed algorithm can effectively detect and extract pavement cracks,which can provide effective information for road disaster detection.