機械学習を用いた画像分析による土壌の粒度分布の迅速評価

抄録

Particle size distribution of soil is often evaluated by “the method for particle size distribution of soils (JIS A 1204)”. This method requires long sieving time and accurate mass measurement of sieved soils, resulting rapid evaluation of in-situ soil characteristics and admission decision of soil classification difficult. Image analysis using an image analyzer, developed by Marui Co. and named “Sand measure”, has been applied for rapid evaluation. The method holds the error generated by the overlaps of soil particles. We have applied machine learning with expanding the input data to image analyzer for reducing the error. The machine learning is installed in the model composed of Convolution Neural Network. The machine learning reduces the error by about 40 % of the results evaluated by test data. Maximum reduction in error is about 78 % in test data. The image analyzer with machine learning can be applicable for rapid evaluation of particle size distribution.