The semi-automated classification of sedimentary organic matter and dinoflagellate cysts in palynological preparations

Palynological material is the entire acid-resistant microscopic organic matter recovered from a sediment or sedimentary rock. This can be subdivided into three broad classes: palynomorphs (i.e., dinoflagellate cysts, spores, pollen, acritarchs, foraminiferal linings, fungal material and algae), palynodebris (i.e., membranous, brown and opaque material) and amorphous OM. The research articulated in this thesis presents a series of integrated procedures to semiautomate the capture, analysis and classification of sedimentary organic matter, with the subsequent clustering of dinoflagellate cysts, in palynological preparations under transmitted-light microscopy. To date, an internationally acceptable classification scheme for sedimentary organic matter has proved illusive. Three recent classification schemes proposed by Boulter (1994), Tyson (1995) and Batten (1996a) have been examined to extract the most suitable set of morphological and textural discriminatory features that are customarily used by palynologists. This has assisted in the development of a classification scheme for use in automated transmitted-light microscopy. In the system developed, the morphological and textural discriminatory features obtained from the classification schemes are related to image analysis algorithms along with several other common algorithms used in classification. By means of a slide capture and coordinate log step, these are then measured for the particles using a computer-controlled stage and a digital camera mounted on a microscope in combination with the Halcon software suite. The Exhaustive CHAID classification tree algorithm is then applied to all feature measurements to establish their saliency as classification discriminators. The results of the classification tree algorithm are then used to determine the databases used by the actual classifier, which consists of a series of Artificial Neural Networks (ANNs). The Gamma Test (GT) is used to help facilitate the best use of limited data and to ensure that the ANNs are not over trained. The palynomorph database created with the greatest amount of salient features is then used with an unsupervised ANN clustering algorithm to predict groups within the dinoflagellate cysts. The results demonstrate that the system developed is able to achieve an average correct classification rate of >90%. Finally, based on a suboptimal feature set, the unsupervised clustering ANN has demonstrated sufficient variation to five distinct clusters within the dinoflagellate cysts, demonstrating the open-ended nature of the system by enabling further subclassification of this group. This is encouraging enough to prompt further research that could result in a commercially viable system.