Characterizing coastal sand composition using image-based machine learning techniques

Authors: Molly Elizabeth Smith*, Florida Atlantic University, Caiyun Zhang, Florida Atlantic University, Anton Oleinik, Florida Atlantic University
Topics: Coastal and Marine, Marine and Coastal Resources, Remote Sensing
Keywords: coastal geology; machine learning; object-based image analysis; OBIA; sand analysis
Session Type: Guided Poster
Day: 4/5/2019
Start / End Time: 3:05 PM / 4:45 PM
Room: Roosevelt 3.5, Marriott, Exhibition Level
Presentation File: No File Uploaded


Sand compositional analysis, which provides information of source materials and depositional environment, is an important component of coastal geology. Beach sand composition influences geomorphology and is an important determining factor to beach width and stability as well as its overall resistance to erosion. Geological-based methods such as microscopic and sieving techniques are moderately fast to complete, but the results from these methods are susceptible to user interpretation. More advanced techniques such as x-ray diffraction/fluorescence are commonly used for sand compositional analysis; and though the results are accurate, such techniques are time-consuming, costly, and labor-intensive. Image-based methods have been proposed for mineralogical analysis, but these studies have largely ignored investigation of sand. In this study, for the first time, we developed an image-based machine learning approach as an alternative to previous sand analysis methods. We collected microscope imagery of coastal and inland sand samples with a varying composition, and applied object-based image analysis techniques and four machine learning classifiers to estimate sand compositions including k-Nearest Neighbor (k-NN), Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF). Accuracy of our approach was validated through comparison with results generated from traditional geological methods in sand compositional analysis. Preliminary results suggest that the image-based method, including image segmentation, demonstrates a very high potential in sand analysis in terms of its analysis speed, reduced subjectivity, and simple semi-automated implementation.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login