Visualization-Guided Classification of Carbonized Seeds from Early Human Civilizations

Information

Abstract

Since the Neolithic Revolution approximately 10.000 years ago, crop plants are an important part of our food. Researchers of archeobotany try to find and determine the species that humankind used already in the past. Most of the gathered samples are preserved due to carbonization, but the shape and inner structures are deformed because of this process. The amount of distortion is given by the temperature and the time they are heated. Normally, an expert is consulted to classify them. Since there are only a few experts in this field, an automatic approach is requested. The result of this work is a software, which can load the Computed Tomography (CT) scans, segment and separate the seeds within the samples, calculate different shape features as descriptors, and train a classifier. To have an overview of how the seeds look like, different volume visualizations are available to show selected samples or median seeds of each class. To validate the probabilities of the learner, additional visualizations are available, which show the influence of the extracted features on the classification. A cross validation method with 1043 known samples results in a classification accuracy of 85 %. The incorrectly classified samples of the ground truth are visualized to display the expert user where they are located regard to the extracted features and which results are especially inaccurate. It turned out, that the opportunity to export the features into a tabular filetype and the visualization of the output probabilities of the classifier for each species were particularly helpful for the domain experts.

Additional Files and Images

Additional images and videos

Additional files

Weblinks

No further information available.

BibTeX

@mastersthesis{Gogel2020,
  title =      "Visualization-Guided  Classification of Carbonized  Seeds
               from Early Human  Civilizations",
  author =     "Andreas Gogel",
  year =       "2020",
  abstract =   "Since the Neolithic Revolution approximately 10.000 years
               ago, crop plants are an important part of our food.
               Researchers of archeobotany try to find and determine the
               species that humankind used already in the past. Most of the
               gathered samples are preserved due to carbonization, but the
               shape and inner structures are deformed because of this
               process. The amount of distortion is given by the
               temperature and the time they are heated. Normally, an
               expert is consulted to classify them. Since there are only a
               few experts in this field, an automatic approach is
               requested. The result of this work is a software, which can
               load the Computed Tomography (CT) scans, segment and
               separate the seeds within the samples, calculate different
               shape features as descriptors, and train a classifier. To
               have an overview of how the seeds look like, different
               volume visualizations are available to show selected samples
               or median seeds of each class. To validate the probabilities
               of the learner, additional visualizations are available,
               which show the influence of the extracted features on the
               classification. A cross validation method with 1043 known
               samples results in a classification accuracy of 85 %. The
               incorrectly classified samples of the ground truth are
               visualized to display the expert user where they are located
               regard to the extracted features and which results are
               especially inaccurate. It turned out, that the opportunity
               to export the features into a tabular filetype and the
               visualization of the output probabilities of the classifier
               for each species were particularly helpful for the domain
               experts.",
  month =      nov,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Research Unit of Computer Graphics, Institute of Visual
               Computing and Human-Centered Technology, Faculty of
               Informatics, TU Wien ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2020/Gogel2020/",
}