Discrete Aggregate Mass Calculation Method for Visual Detection of Aggregate Gradation and Elongated and Flat Aggregate Contents

Aggregate gradation and elongated and flat aggregate contents strongly affect the performance of asphalt mixtures. During the visual detection of these two indexes, the morphology in a single view is typically used for mass calculation. However, it has a significant error and affects the detection accuracy. Therefore, in this study, the morphologies of an aggregate from multiple views were collected during falling. Size features were also extracted for mass calculations. Two mass calculation methods, the multi-view equivalent volume models (MEVMs) and ensemble regression learning model (ERLM), were proposed in this study. MEVMs were constructed using multi-view shape features. The relationships between pixel volumes of MEVMs and actual aggregate mass were established through the least square method for mass calculations. Correlation analyses of multi-view size features were conducted and weakly correlated features were eliminated. The ERLM was combined with the K-nearest neighbor algorithm, multi-layer perceptron neural network, support vector regression algorithm, and ensemble decision tree using an adaptive weight assignment algorithm. The ERLM was trained with processed multi-view features for mass calculations. Finally, the feasibility of MEVMs and ERLM were verified through mass calculations of aggregates with different particle sizes and shapes. Both methods showed significantly improved correlation and accuracy, with the ERLM showing stronger generalization ability in particle size and shape scales than that of MEVMs. Therefore, the ERLM could be effectively applied for the visual detection of aggregate gradation and elongated and flat aggregate contents. The application of the proposed methods was verified in practical road engineering.