FEASIBILITY STUDY ON SOIL CLASSIFICATION FROM SOIL IMAGES USING DEEP LEARNING

Abstract

In this paper we propose an AI-based method for soil classification from images of soil particles. The AI model was trained by virtual soil particle images created on a computer. Under the condition that soil particles were not in contact with each other, one hundred images of real crushed stones with different grain size distribution were classified into simulated seven categories and the 78% estimate was correct. Although further improvement is needed for practical use, this feasibility study showed the possibility of substituting computer-generated images for training data of actual soils.