基于三维纹理特征的沥青路面级配预估方法

Aggregates Gradation Estimation Based on Three-Dimensional Texture Feature Extraction

摘要:

采用三维激光扫描获取路面纹理数据，提取基于几何特征的表面性状参数和基于二维小波变换的多尺度特征参数作为模型输入。通过对比参数和分类器对8种已知级配沥青车辙板的分类效果，验证了表面性状参数与二维小波参数相结合，在多层感知机分类下效果最佳。选用多层感知机多元回归算法对8种路面类型级配曲线进行预估，其拟合优度R²高达0.849。采用消融实验分析不同参数的贡献度，发现二维小波能量参数影响较为显著。

Abstract:

In this paper， a 3D （three-dimensional） laser scanner was used to obtain texture data. Surface geometric parameters and 2D-wavelet-based multiscale parameters were proposed as inputs. Different parameters and classifiers were tested on eight types of mixtures with the standard gradation curve. The result shows that the MLP （multi-layered perceptron） performs best with the combination of the surface and 2D-wavelet parameters. The MLP regression model was used to estimate the gradation curves， and the R² was 0.849. The ablation experiments were used to analyze the contribution of different parameters. It is found that the 2D-wavelet energy parameters have more significant effects.