DIA for Classification of Soils Using Machine Learning and Computer Vision

This chapter demonstrates that computer visionVision or MachineMachine LearningLearn (ML) combined with dataData collected using DIA permits classification of individual sand particles with suitable accuracyAccuracy. ML modelsModel and other Artificial IntelligenceArtificial intelligence (AI) (AI) methods are now ubiquitous in ecommerce and are being used extensively in everyday interactions to eliminate repetitive tasks previously carried out by humans. However, trainingTrain of ML and AIArtificial intelligence (AI) modelsModel requires large data setsDataset that are typically not available through traditional methods of acquiring geotechnical dataData. DIA provides a large number of images that can be used for ML based sand classification. This chapter explores how DIA can be leveraged for automatingAutomation particle classification of sand. In particular, the use of size and shape features acquired using DIA for particle classification is presented. Both individual and ensembleEnsemble machineMachine learningLearn modelsModel were employed, with several modelsModel yielding promising results. In addition, images captured from DIA are also incorporated with size and shape descriptors for sand classification. A ScaleScale InvariantInvariant Feature TransformTransform (SIFT) wasSIFT adopted to eliminate scaleScale effects in the acquired images. The importance and efficacy of using size and shape features was compared to using SIFTSIFT features alone. It was determined that SIFTSIFT can improve the accuracyAccuracy of ML modelsModel, but use of size and shape parameters was key to modelModel accuracyAccuracy.