Development of AI Model and App to estimate soil grain size distribution from photographic images

Abstract

In this study, a grain size analysis AI combining image analysis and deep learning was developed to quickly and accurately estimate grain size distribution, which is important in geotechnical engineering. As a labor-saving alternative to JIS standard tests, a CNN was constructed and improved using approximately 40,000 actual measurement data and captured images, achieving an estimation accuracy of ±5%. Field trials at construction sites confirmed significant time savings and high practicality compared to conventional methods. The system is available as a web application, and expanding the measurement range is a future challenge.