Three‐dimensional quantitative analysis on granular particle shape using convolutional neural network

Abstract To identify all desired shape parameters of granular particles with less computational cost, this study proposes a three‐dimensional convolutional neural network (3D‐CNN) based model. Datasets are made 100 ballast and Fujian sand particles, the (i.e., aspect ratio, roundness, sphericity, convexity) obtained by conventional methods used to label particles. For model training, feeding slice images into model, contour is automatically extracted, thereby can be learned Thereafter, applied predict new as testing. All results indicate trained on cut from three orthogonal planes presents highest prediction accuracy an error than 10%. Meanwhile, for concave angular guaranteed. The rotation‐equivariant confirmed, in which predicted values roughly independent orientations particle when cutting images. Superior methods, desirable one unified 3D‐CNN its complexity number triangular facets, thus saving computation cost.