一种利用骨架提取和SVM分类的颗粒表征方法

集料级配是影响沥青路面抗滑和减噪性能的最主要因素，基于机器视觉技术，提出一种集料粒径高精度无损检测方法，结合骨架提取和支持向量机（SVM）算法来实现集料颗粒粒度分布的精确估计，为保证集料级配质量提供了有效手段。集料颗粒图像由数字相机获取，图像标定基于NI视觉助手Vision Assistant软件完成。颗粒骨架图像采用Skeleton-M骨架提取算法提取，针对所出现的骨架断裂情况，使用形态学膨胀和细化算法完成骨架修复；然后基于颗粒骨架图像提出颗粒粒径表征算法，基于骨架端点个数提出颗粒棱角表征方法；最后，通过提取颗粒图像特征参数，使用SVM算法对颗粒粒径分布进行精确估计。在颗粒棱角性表征方面，将所提出的颗粒骨架棱角性表征法与AIMS系统的梯度棱角法、未压实空隙率集料颗粒棱角法进行了比较。结果表明：与目前常见的粒径表征方法相比，所提出的方法可以较好地实现颗粒粒径表征和颗粒棱角性表征，所采用的基于多特征的SVM颗粒分档算法可以有效实现颗粒粒径分档，筛分档间的分档精度最高可达95%以上。

Method for Particle Characterization by Using Skeleton Extraction and SVM Classification

Measuring the particle size with high accuracy and efficiency is important for ensuring the quality of a pavement construction and directly affects the long-term performance of the pavement. This paper proposed a new approach for evaluating the distribution of the particle size by combining the skeleton extraction and support vector machine (SVM) algorithms. The particle images were captured using a design image acquisition system. The skeleton images were extracted under the NI environment following the image segmentation process. Next, the skeleton repairing algorithm was designed to repair the fractured skeleton images. Finally, the distribution of particle size was estimated by employing the SVM method. The aggregate angularity was evaluated based on the final number of the skeleton images. The correlation analysis and comparison were conducted by the AIMS and uncompacted void fraction aggregate particle angular method. When compared with the particle size characterization method, the skeleton extraction method not only identifies the characteristics of the aggregate particle size but also evaluates the aggregate angularity. Because of limited characterization precision of the particle size, the nonlinear SVM with kernel rbf can distinguish the particle sieve size effectively, and the characterization precision can be estimated up to 95% accuracy.