Laboratory, Field and In-Situ Soil Characterization Through Image Processing

点击翻译按钮获取中文摘要

Soil classification is the first and foremost task of geotechnical, geoenvironmental and geotechnical earthquake engineering site characterization. Assessment of grain size distribution is essential to the classification. Despite significant shortcomings, grain size analysis is ubiquitously performed by sieve and hydrometer tests. A research program is therefore being undertaken to develop the hardware, analytical tools and software for characterization of soils from images of grain assemblies collected in the laboratory; above ground in the field; and from below the ground surface. The effort may result in cleaner, quicker, more efficient, less costly and more accurate grain size determination and stratigraphic delineation. Earlier image processing efforts were successful at determining the dominant grain size when particles are relatively uniform in size, or if the particles can be quickly segregated by size prior to image capture. The assessment of grain size distribution from images becomes orders of magnitude more complex when a soil contains a range of grain sizes. Previous techniques using image textural features and wavelet mathematics have paved the way for new methods including edge pixel density (EPD), edge segment size (ESS) and mathematical morphology that will yield more complex grain size distributions. The newer techniques may necessitate image capture at a range of magnifications.For the laboratory, a system is being developed for rapidly segregating a soil specimen by size using a sedimentation column. This permits piecewise image collection and assemblage of a traditional grain size distribution curve. The hardware and image collection system is being designed to facilitate size recognition down to 0.005 mm diameter, the commonly cited size threshold between ?silt? and ?clay?. The hardware development will be performed as part of an interdisciplinary undergraduate design project with students from several disciplines including civil/geotechnical engineering, mechanical design, image processing and mathematical statistics.For field applications, the research will develop a rapid response system to determine the grain size distribution of soils at the ground surface using two different approaches: mathematical morphology and edge detection to determine edge pixel density (EPD) and edge segment size (ESS). Preliminary evidence suggests these methods applied to images taken at multiple scales may provide the grain size distribution of even well-graded soils without the need for particle segregation. The research will also develop a rapidly deployable field portable computer and camera system. For subsurface grain size assessment, the vision cone penetrometer (VisCPT) will be used for image collection. Earlier studies have shown that textural indices can only detect changes in stratigraphy, but grain size determination has been too complex a problem. With the newly proposed image processing methods, in-situ grain size determination becomes tractable. VisCPT data will be collected at an exceptionally well characterized sand and gravel quarry in southwestern Indiana. Existing available date data includes continuous soil sampling and grain size distributions to a depth of 20 meters. The glacial outwash site includes a large assortment of soil gradations ranging from sandy gravels to clayey silts.The imaging methods developed herein for soil grain size characterization have direct application to other disciplines and industries where the sizing of manufactured components at various scales, ingredients in preparations, naturally occurring biologic or geologic matter, and interpretation of remotely sensed objects through image processing are essential to quality control, inspection, diagnosis, state assessment, and prediction of future behavior. Such industries include pharmaceuticals, food processing, materials science, powder metallurgy, microbiology and others.