Determination of Soil Grain Size Distribution by Soil Sedimentation and Image Processing.

Abstract

Over the last decades, advances in digital imaging have moved many industries to adopt digital image processing (DIP) to determine the grain size distribution of their products. However, despite of the fact that image-based soil grain size assessment methods have advantages over traditional sieve testing, they have lagged behind due to the inability of single camera-lens systems to capture the wide range of soil grain sizes. Since no DIP technique exists for determination of grain size distribution of 3-dimensional soil assemblies of non-uniform sized particles, an eight foot long glass sedimentation column was constructed to rapidly segregate particles by size prior to imaging. Following sedimentation through the water-filled column a camera is used to collect digital images of the segregated soil sediment. Image sections with a height of 256 pixels, which contain relatively uniform sized particles, are image-processed for soil grain size to obtain volume-based soil grain size distribution. With this approach, the need for determining soil grain sizes from images of non-uniform sized grains is eliminated. The method is termed ‘Sedimaging’. To determine soil grain size at each image section, statistical DIP methods based on wavelet decomposition, pattern spectrum, and edge pixel density were developed. They utilize a ‘wavelet index’ (CA), the ‘edge pixel density’ (EPD), and a ‘structuring element size at peak of pattern spectrum’ (SP) calibrated against the soil grain size in units of pixels per diameter. The effects of effective stress and soil grain size on void ratio in a sedimented soil column were studied to address the influence of void ratio variations on soil grain size distribution by Sedimaging. Void ratio variations developed in the soil sediment were found to be so small that no correction to the image-based grain size distribution is necessary. Soil grain size distribution by Sedimaging fairly well mimics the sieve-based grain size distribution. In particular, the wavelet decomposition and pattern spectrum methods demonstrated their suitability to Sedimaging. However, the edge pixel density method’s implementation into Sedimaging was not as successful as the other two methods mainly due to the sensitivity of EPD to grain size uniformity.