Quantification of soil features using digital image processing (DIP) techniques

A new thin section method is introduced which provides reliable, automated classification of mineral, non-mineral constituents (e.g. organic matter), non-crystalline, or poorly crystalline components (e.g. Fe–Mn oxides) and voids. A color image flatbed scanner scanned 10 soil thin section slides that contain the same features. Equal portions (about 6.3 cm2) of each slide were imported into the Erdas Image Processing software (version 8.4) as 24 bit 3-band images. Images were classified with an unsupervised nearest neighbor classification method with several different processing steps. Five different classes were separated and quantified for each sample. Classified features were checked with 500 reference points under the petrographic microscope. Separation and identification was almost 100% for calcite, about 97% for void in all samples, but values decreased for sesquioxides, plasma, and quartz (96%, 96%, and 80%, respectively). Quantitative results of digital image processing based on pixel number of each class (aerial percentage) were compared with traditional point-counting method. Digital image processing results showed slightly lower values for voids, quartz, and sesquioxides, but higher values for plasma and almost equal quantity for calcite in all three samples when compared with the values of the point-counting method. This technique represents a significant improvement in quantitative soil micromorphology. Requirement of simple and inexpensive hardware and quick and routine identification and quantification of features (calcite, void, sesquioxides, and plasma) with much less error than other methods are two advantages of the proposed method to the earlier studies.