Fish otoliths have long played an important role in sustainable fisheries management. Stock assessment models currently used rely on species specific age profiles obtained from the seasonal patterns of growth marks that otoliths exhibit. We compare methods widely used in fisheries science (elliptical Fourier) with an industry standardised encoding method (MPEG7 - Curvature-Scale-Space) and with a recent addition to shape modelling techniques (time-series shapelets) to determine which performs best. An investigation is carried out into transform methods that retain size-information, and whether the boundary encoding method is impacted be otolith age, performing tests over three 2-class otolith datasets across six discrete and concurrent age groups. Impact of segmentation methods are assessed to determine whether automated or expert segmented methods of boundary extraction are more advantageous, and whether constructed classifiers can be used at different institutions. Tests show that neither time-series shaplets nor Curvature-Scale-Space methods offer any real advantage over Fourier transform methods given mixed age datasets. However, we show that size indices are most indicative of fisheries stock in younger single-age datasets, with shape holding more discriminatory potential in older samples. Whilst commonly used Fourier transform methods generally return best results; we show that classification of otolith boundaries is impacted by the method of boundary segmentation. Hand traced boundaries produce classifiers more robust to test data segmentation methods and are more suited to distributed classifiers. Additionally we present a proof of concept study showing that high energy synchrotron scans are a new, non-invasive method of modelling internal otolith structure, allowing comparison of slices along near infinite numbers of virtual complex planes.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:687883 |
| Date | January 2015 |
| Creators | Mapp, J. |
| Publisher | University of East Anglia |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | https://ueaeprints.uea.ac.uk/59388/ |
Page generated in 0.0015 seconds