摘要 : Particle angularity significantly affects the macro-mechanical behavior of granular soils. However, due to the difficulty of characterizing particle angularity, this fundamental soil property is commonly ignored by researchers and... 展开 Particle angularity significantly affects the macro-mechanical behavior of granular soils. However, due to the difficulty of characterizing particle angularity, this fundamental soil property is commonly ignored by researchers and practitioners in geotechnical applications. This study develops a smartphone application allowing the automatic evaluation of the particle angularities of soils. Therefore, this technique is termed as laboratory-on-a-smartphone. A total of 75,000 various granular soil images are collected in this study. Based on their roundnesses, these images are labeled into six classes including very angular, angular, subangular, subrounded, rounded, well-rounded soils, following the Powers' chart. Then, machine learning techniques, including speed up robust features, k-means, and support vector machine, are used to train a soil image classifier. This soil image classifier automatically analyzes the sharpnesses of particle corners in three-dimensional soil assembly images and classifies images based on Powers' chart with a high classification accuracy of 93%. This technique does not require a specialized device to capture images other than a smartphone. It can achieve real-time angularity evaluations without demanding computations. It is fully automated without human intervention. These features ensure that researchers and practitioners can easily implement this technique in the field and laboratory. 收起