基于多特征流形学习和矩阵分解的路面裂缝检测

针对单一属性特征的路面裂缝检测方法无法从复杂背景噪声中准确提取裂缝信息的缺陷,提出一种结合多特征流形学习和矩阵分解的路面裂缝检测算法。该算法首先根据路面裂缝子块的统计、形状和纹理特性抽取多重属性特征并构造多个流形正则项,将流形正则项嵌入于矩阵分解的目标函数中,采用交替迭代法在统一框架下实现裂缝子块降维和多特征自适应融合。为进一步提高聚类准确率,对路面裂缝图像采用各向异性算法增强得到少量有效样本标签,实现算法的半监督扩展。在公开数据集(Crack IT)和实际采集的沪宁高速(HN)路面图像库上的实验结果表明,该算法可以有效提高路面裂缝识别率,验证了算法的有效性。

Pavement crack detection based on multi feature manifold learning and matrix factorization

Aiming at the defect that using only single type of attribute feature, conventional pavement crack detection methods cannot accurately extract crack information from complicated background noises; in this paper, a novel pavement crack detection algorithm is proposed based on integrating multi feature manifold learning and matrix factorization. Firstly, according to the statistics, shape and texture features of the pavement crack sub patches, the multiple attribute features are extracted and multiple manifold regularized terms are established. Then, the manifold regularized terms are embedded into the objective function of matrix factorization. Finally, an alternating iteration algorithm is adopted to realize the dimension reduction of the crack sub patches and adaptive multiple feature fusion within a unified framework. To further improve the clustering accuracy, an anisotropy algorithm is adopted to enhance the pavement crack image, partial effective sample labels are obtained, and the semi supervised expansion of the algorithm is achieved. The experiment results on the public pavement crack dataset (CrackIT) and the practically acquired Hu Ning (HN) highway pavement crack image database show that the proposed algorithm can effectively improve the pavement crack recognition rate, which verifies the effectiveness of the algorithm.