Schoening, Timm: Automated detection in benthic images for megafauna classification and marine resource exploration: supervised and unsupervised methods for classification [...]. 2015
Inhalt
- Abstract
- Publications
- Acknowledgements
- Contents
- List of Tables
- List of Figures
- Acronyms
- Principles and Background
- 1 Introduction
- 1.1 Ocean exploration
- 1.2 Curse of dimension
- 1.3 Adding semantics
- 1.4 Computer Vision for the Deep Sea
- 1.5 Scope
- 1.6 Contributions
- 1.7 Notation
- 1.8 Overview
- 2 Benthic imaging
- 2.1 Video acquisition
- 2.2 Camera platforms
- 2.2.1 Towed systems
- 2.2.2 Autonomous Underwater Vehicles
- 2.2.3 Remotely Operated Vehicles
- 2.2.4 Lander and Crawler
- 2.2.5 Others
- 2.3 Light and colour
- 2.4 Quantification of image content
- 2.5 Filtering for analysable images
- 3 Pattern Recognition
- 3.1 Digital Images
- 3.2 Feature Descriptors
- 3.2.1 Colour and intensity
- 3.2.2 Histograms
- 3.2.3 Gabor wavelets
- 3.2.4 MPEG-7 descriptors
- 3.2.5 Blob descriptor
- 3.2.6 SIFT/SURF
- 3.3 Feature metrics
- 3.4 Feature normalisation
- 3.5 Feature selection
- 3.6 Unsupervised Machine Learning
- 3.6.1 k-Means
- 3.6.2 Self-Organising Map
- 3.6.3 Hyperbolic Self-Organising Map
- 3.6.4 Hierarchical Hyperbolic Self-Organising Map
- 3.7 Supervised Learning
- 3.8 Other methods
- 3.9 Quality criteria
- 3.10 Training data division and parameter tuning
- 4 Annotation
- Scenarios and Contributions
- 5 Colour normalisation
- 6 Laserpoint Detection
- 7 Megafauna detection
- 7.1 Initial dataset
- 7.2 Semi-automatic detection of megafauna
- 7.2.1 Manual annotation of POIs with BIIGLE
- 7.2.2 Creation of annotations cliques
- 7.2.3 Colour pre-processing with fSpice
- 7.2.4 Feature extraction at POIs
- 7.2.5 Feature extraction in a ROI
- 7.2.6 Feature normalisation
- 7.2.7 Training set generation from cliques
- 7.2.8 SVM trainings and parameter tunings
- 7.2.9 Classification of ROI features with SVMs
- 7.2.10 Post-processing to derive detection positions
- 7.3 Results
- 7.4 Re-evaluation
- 7.5 Multi-year assessment
- 7.6 Other methods
- 7.7 Other data sets
- 8 Benthic resource exploration
- Outlook
- 9 Ideas for the Future
- 9.1 Further methods
- 9.1.1 Image normalisation
- 9.1.2 Feature Descriptors
- 9.1.3 Post-processing
- 9.1.4 Black boxes
- 9.1.5 Imaging hardware
- 9.2 Further annotation ideas
- 9.3 Marine applications
- 9.4 Integrated visual programming
- 10 Conclusion
- Appendix
- A Further images and visualisations
- B Rapid development of high-throughput methods
- B.1 The idea behind Olymp
- B.2 Infrastructure
- B.3 Basic libraries and tools
- B.4 Pan
- B.4.1 Zeus
- B.4.2 Atlas
- B.4.3 Plutos nodule browser
- B.4.4 Poseidon
- B.4.5 Ate
- B.4.6 tinySQL
- B.4.7 Spectra
- B.4.8 Delphi
- B.5 High-throughput
- B.6 Outlook
- Bibliography
- Declaration