Nonlinear Kernel Convolutional Neural Network to Find Median Sand Particle Size

• Published in: Journal of coastal research 2021-10, Vol.114 (sp1), p.1-5
• Main Authors: Kim, Hyoseob, Yoo, Ho Jun, Lee, Jung Lyul
• Format: Article
• Language: English
• Subjects: Artificial neural networks; CNN; COASTAL ENGINEERING; Coastal inlets; Coastal management; Coastal research; Coefficients; Convolution; Feature maps; image recognition; Kernels; Mean sand particle size; Neural networks; Particle size; Particle size distribution; Sand; Sand & gravel; sediment particle; Size distribution; SNN; Subtraction
• ISSN: 0749-0208; 1551-5036
• DOI: 10.2112/JCR-SI114-001.1

Kim, H.; Yoo, H.J., and Lee, J.L., 2021. Nonlinear Kernel Convolutional Neural Network to find median sand particle size. In: Lee, J.L.; Suh, K.-S.; Lee, B.; Shin, S., and Lee, J. (eds.), Crisis and Integrated Management for Coastal and Marine Safety. Journal of Coastal Research, Special Issue No. 114, pp. 1-5. Coconut Creek (Florida), ISSN 0749-0208. Convolutional Neural Network (CNN) has successfully been used in various areas. We focus on predicting various sand particle sizes by applying convolutional neural network with a nonlinear kernel. The nonlinear kernel involves a bias and negative square of subtraction between input image pixel numbers and the kernel coefficients and summation. The convolution layer conform new feature map in Convolutional Neural Network. While using batch gradient descent method to train relevant coefficients and biases, the gradient of the square of subtraction term appears in the whole gradient over each kernel coefficient. The network was examined on regular-sized sands, i.e. 2000, 1000, 500, 250, 125 and 63 micrometer. The network was trained by using various images for each size. It was validated against new images and the absolute error was less than 30 micrometer, respectively, which is satisfactory. The network was applied by using 3 images of sands with size distribution. The results show good validation and satisfactory predictions. In the course of study, several numbers of kernels, kernel sizes, pooling sizes were tried and the optimum architecture for this work was chosen. It is expected that the present network will reduce time and effort in obtaining median sand size in many field projects. The size distribution of sand particles could also be obtained with the present network in the near future.