Method for identifying and dividing aggregate boundary in asphalt mixture based on deep learning

Abstract

The invention discloses a method for identifying and dividing aggregate boundary in asphalt mixture based on deep learning, and the method comprises the steps of firstly constructing an asphalt mixture cross section image data set, and employing a customized multi-threshold binarization method to increase the image marking precision in the data set construction process; then, building an asphalt mixture CT section segmentation neural network, and carrying out further segmentation on a neural network image segmentation result in combination with a watershed algorithm; and finally, extracting aggregate mesoscopic information in the segmented image. According to the method, the rapid and efficient acquisition process of aggregate microstructure information in the asphalt mixture is realized, the neural network is beneficial to more accurate and intelligent image segmentation, and the improvement of the CT image segmentation precision of the asphalt mixture is further realized by introducing a data set labeling method and a watershed algorithm, and convenience is provided for obtaining the mesoscopic information of the aggregate in the asphalt mixture.

Description

technical field [0001] The invention relates to a method for identifying and dividing aggregate boundaries in asphalt mixtures based on deep learning, and belongs to the technical fields of computer vision technology and road engineering materials. Background technique [0002] Road engineering plays a pivotal role in the development of the national economy, and asphalt pavement is widely used in road engineering based on its smoothness and comfort. As the main material of asphalt pavement, the performance of asphalt mixture is greatly affected by the internal microstructure. Asphalt mixture is mainly composed of aggregates, cement and voids. Cutting and industrial CT scanning are often used to obtain the internal structure of asphalt mixture. Among them, industrial CT is favored by scholars for its advantages in non-destructive testing. In order to obtain the mesoscopic information of aggregates such as area and centroid in CT scanning images, it is particularly important ...

AI Technical Summary

Problems solved by technology

[0003] (1) Segmentation using a single threshold such as OTSU. In practice, the grayscale of each part of the image is uneven. Using a single threshold often cannot effectively segment the image, and there will be a large number of aggregates sticking together and unrecognized aggregates.

[0004] (2) Using the method of multi-threshold segmentation, this method often cuts the image according to certain rules, finds the best threshold for each cut and then performs image segmentation, which can suppress the occurrence of a large number of adhesions and unrecognized phenomena, but the recognition There are many noises inside the aggregate, and further processing is required between clipping and clipping to ensure coherence

[0005] (3) The segmentation method of edge recognition is adopted. This kind of method retrieves the image edge according to the image gradient, such as Canny operator edge recognition. The edge detection based on the gradient can identify the contact edge between the aggregate and the mortar, but it is easy to regard the noise as oversegmentation

Images