Aggregate particle identification and gradation analysis method based on the deep learning network of Mask R-CNN

Abstract

Aggregate gradation detection is an indispensable work for building materials. However, sieving test is time-consuming and labor-intensive, and the analysis accuracy of the current image analysis methods is limited. Accordingly, this paper aims to establish an efficient and accurate aggregate particle identification and aggregate gradation analysis method based the network of mask R-CNN. First, the aggregate particle identification model was trained and evaluated, and the robustness of the model was discussed. Afterwards, the aggregate gradation analysis method was proposed and verified. Transfer learning was employed to enable the model to identify other types of aggregate. The results show that the aggregate particle identification model established in this paper could accurately identify aggregate particles, and it showed excellent robustness to mud content. The equivalent aggregate particle size determined by differentiating the aggregate into an ellipse or a circle based on the ratio of the length and width of the minimum external rectangle of the aggregate image was intensely close to the ground truth. The gradation analysis results by the method proposed in this paper were extremely close to the sieving results. After transfer learning, the missing identification and false identification were significantly improved, the model could accurately identify various types of aggregates simultaneously.