Abstract
Particle size distribution (PSD) detection is an important part of the construction process, which may delay subsequent construction if it takes too much time. Image-based methods have been proposed for effective and inexpensive PSD detection based on digital image processing (DIP) or deep learning. However, current image-based PSD detection methods mostly ignore the occluded regions of gravel particles, which causes errors in the detection process. Hence, this study proposed a novel occlusion-aware PSD detection method, which adapted an amodal instance segmentation model, improved Bilayer Convolutional Network (BCNet) with an Optimized Feature Pyramid Network (FPN), and a backbone embedding with global context network (GCNet) module, for occlusion-aware particle image segmentation. The improved BCNet has achieved 6.7% AP50 and 6% AP75 improvement compared to the original BCNet. For the results of PSD analysis, the maximum and average absolute errors of the mass fraction of various particle size intervals were 3.38% and 1.27%, respectively, less than 4.65% and 2.05% of the way only considering visible regions of particles, which indicates the proposed method can reduce the error of image-based PSD detection method only considering visible regions of particles. Furthermore, the detection time of the proposed method is less than 120 s which is more rapid than mechanical sieving.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call