Final presentation for Ordinance Survey sponsored MSc Project

2,676 views

Published on

MSc Archaeological Computing (GIS and Survey), University of Southampton.
“An archaeological reaction to the remote sensing data explosion. Reviewing the research on semi-automated pattern recognition and assessing the potential to integrate artificial intelligence”

Published in: Education
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
2,676
On SlideShare
0
From Embeds
0
Number of Embeds
2,323
Actions
Shares
0
Downloads
5
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • First features could be identified which could not be seen, or at least appreciated fully, on the ground. O.S.G. Crawford, first archaeological officer of the British Ordnance Survey.
  • First features could be identified which could not be seen, or at least appreciated fully, on the ground. O.S.G. Crawford, first archaeological officer of the British Ordnance Survey.
  • Increasing availability, extent and lower prices of VHR remote sensing
  • Shift in scope - Geosciences, Multi-sensor, Detect variability, Large areas. Artificial intelligence (AI), Replicate humans
  • Although archaeologists have attempted to formalise arguments and created computer systems to critically reflect arguments, the creation of successful AI systems are very rare. The main problems in rule definition within the archaeological record is that it is almost always incomplete: not all past material things have remained until today or are found destructed. So, in order to replicate an archaeologist it must be able to reconstruct incomplete data for which Barceló— (2008, 49) identified the following key concepts.
  • Image - (left) original image, (middle) fusion of intensity and texture gradient images, (right) segmentation results
  • Image - A basic, three-layer neural network topology, with a hidden layer
  • An illustration of the architecture of the CNN, explicitly showing the delineation of responsibilities between the two GPUs. One GPU runs the layer-parts at the top of the figure while the other runs the layer-parts at the bottom. The GPUs communicate only at certain layers
  • Result of the image segmentation into objects. Colours outlining the objects display the range of brightness with highest values in green (location: Overton Hill).
  • Increasing availability, extent and lower prices of VHR remote sensing
  • Increasing availability, extent and lower prices of VHR remote sensing
  • Learning how parts of features can be related to a classification (e.g. circular mound surrounded with ditch and bank is a round barrow)
  • Final presentation for Ordinance Survey sponsored MSc Project

    1. 1. An archaeological reaction to the remote sensing data explosion. Reviewing the research on semi-automated pattern recognition and assessing the potential to integrate artificial intelligence. Iris Kramer MSc Archaeological Computing (GIS and Survey) External supervisor: David Holland 14 December 2015
    2. 2. 2 Introduction • Aerial survey in Archaeology • Using AI to imitate the archaeologist • Case study: barrow detection using TRIMBLE eCognition • Discussion and future scope • Conclusion • Next steps
    3. 3. Aerial survey in Archaeology
    4. 4. 4 after Lasaponara and Masini (2012) Aerial photography • First features recorded at large scale by O.G.S. Crawford – From 1920’s Possible cause to the presence of crop marks
    5. 5. 5 Challis et al. (2011) Light Detection And Ranging • First demonstrated in a collaboration of the UK Environment Agency and English Heritage around 2000 • Revolutionary for forested areas since 2006 Interaction of laser pulse with forest canopy resulting multi returns over increasing time
    6. 6. 6 Automated methods • Shape detection – e.g. lines, corners, circles • Template matching Rectangularity heath map derived from Hough transform line detections after Zingman et al. (2015) (a) The ground plan and cross-section geometry of a charcoal kiln site. (b) LiDAR derivatives for template matching Schneider et al. (2015)
    7. 7. 7 Reacting to the data explosion • “…there will never be any automated mapping for archaeology…” – Parcak 2009 • “…focus should be on predictable shapes and sizes as these work best within the presented template matching and shape detection algorithms…” – Bennett et al. 2014 • Limited research
    8. 8. Using AI to imitate the archaeologist
    9. 9. 9 • Key concepts for reconstructing stories - Barceló (2008) • Deduction (argumentation) • Induction (learned from examples) • Analogy (information recalled from previous case studies) Archaeological discovery: incomplete data
    10. 10. 10 • Geomorphic fingerprint – Define rules Human argument: cognitive computing after van den Eeckhaut et al. (2012) Process of visual interpretation of archaeological features
    11. 11. 11 Barceló (2008) Human experience: machine learning • Artificial Neural Network • Some examples The basic, three-layer neural network topology, with a hidden layer A neural network to recognize visual textures as use-wear patterns in lithic tools
    12. 12. 12 (top) Barceló 2008, (bottom) Krizhevsky et al. 2012 Human experience: machine learning • Artificial Neural Network • ImageNet contest 2012 – Deep convolutional neural network The CNN architecture, explicitly showing the delineation of tasks between two GPUs. The basic, three-layer neural network topology, with a hidden layer Results of test images and labels found most probable by the model
    13. 13. Case study: barrow detection using TRIMBLE eCognition
    14. 14. 14 Reinvention of eCognition • Not useful for archaeology? • Very useful for landslide detection! de Laet et al. (2007) Result of classifying shadows of walls Overview of processing steps for the Random Forest algorithm Stumpf and Kerle (2011)
    15. 15. 15 Avebury, Wiltshire • Prehistoric landscape • LiDAR data from the Environment Agency – Slope derivative • Aerial photography from OS
    16. 16. 16 Feature detection 1. Defined by rules 2. Template matching 3. Towards automation •Most attempted feature detection – Round barrows Various types of barrows
    17. 17. 17 Defined by rules Three barrow types; (left) Bell (middle) Saucer (right) BowlImage segmentation into objects with range of brightness Open test image Define features Generate threshold Classify features Review classification Add threshold Open verification image Apply ruleset Evaluate result Export classification Image segmentation Define segmentation threshold Iterate process Iterate process Iterate process
    18. 18. 18 Template matching Barrow classification based on correspondence thresholdFive template barrows created from training locations Open test image Sample selection Generate template Test template Define threshold Review targets Update template Open verification image Create correlation map Evaluate correlation Execute classification Iterate process Export classification Iterate process Iterate process
    19. 19. 19 Towards automation Image segmentation into objects trained on brightness Open test image Assign class to test features Train RF classifier Apply RF classifier Open verification image Apply ruleset Review classification Export classification Image segmentation Define segmentation threshold Iterate process Iterate process Open test features
    20. 20. 20 Evaluation • Best results through defined rules • Most potential for self-learning algorithm Other saucer bell bowl True positive 10 3 14 23 False negative 76 7 6 74 Percentage p/n 12% 30% 70% 24%
    21. 21. Discussion and future scope
    22. 22. 22 AI in reaction to the data explosion • Ever increasing data from various sources – “Is satellite technology advancing faster than archaeologists’ ability to learn, apply, and analyse the data and programs, and all the inherent implications?” - Parcak (2009) – Limited research in overall methods • Heritage monitoring • Small scale
    23. 23. 23 • Consistency in large mapping programmes – Exchange of common feature detection • (e.g. ditch, mound) – Web-based data repository Future scope Round barrow Mound Round has shape is defined by … (varied sizes) has size Ditch possibly surrounded by Bank possibly surrounded by Flora Agriculture possibly (partly) levelled Fauna possibly (partly) destroyed has landcover Barrow Earthworkis type of is type of is type of Semantic description of a round barrow
    24. 24. Conclusion
    25. 25. 25 Research in automated feature recognition • Limited in-depth research – Short – On-the-side – No knowledge exchange – Settled for less • Lot of potential – Emerging research in Geosciences and Computer Vision – Reaction to hazards, long term changes, building projects
    26. 26. Next steps
    27. 27. 27 PhD in machine learning? • Creation of a reference database such as ImageNet – 14,197,122 images? – Connecting objects to words • Application for archaeology – Connecting parts of features to words (e.g. ditch, mound) – Deep learning • Multi-scalar • Parts of features related to types
    28. 28. Bibliography Barceló, J. A. 2008. Computational Intelligence in Archaeology, Hershey, New York, IGI. Bennett, R., Cowley, D., and De Laet, V. 2014. The data explosion: tackling the taboo of automatic feature recognition in airborne survey data. Antiquity, 88, 896-905. van den Eeckhaut, M., Kerle, N., Poesen, J., and Herv‡s, J. 2012. Identification of vegetated landslides using only a Lidar-based terrain model and derivatives in an object-oriented environment. Proceedings of the 4th GEOBIA, 211. Krizhevsky, A., Sutskever, I., and Hinton, G. E. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 2012. 1097-1105. Lasaponara, R., and Masini, N. 2012. Image Enhancement, Feature Extraction and Geospatial Analysis in an Archaeological Perspective. In: Lasaponara, R., and Masini, N. (eds.) Satellite Remote Sensing: a New Tool for Archaeology. New York: Springer. de Laet, V., Paulissen, E., and Waelkens, M. 2007. Methods for the extraction of archaeological features from very high-resolution Ikonos-2 remote sensing imagery, Hisar (southwest Turkey). Journal of Archaeological Science, 34, 830-841. Niemeyer, I., Marpu, P. R., and Nussbaum, S. 2008. Change detection using object features. In: Blaschke, T., Lang, S., and Hay, G. J. (eds.) Object-Based Image Analysis: Spatial Concepts for Knowledge-Driven Remote Sensing Applications. Verlag: Springer. Parcak, S. 2009. Satellite Remote Sensing for Archaeology, New York, Taylor & Francis. Schneider, A., Takla, M., Nicolay, A., Raab, A., and Raab, T. 2015. A Template-matching Approach Combining Morphometric Variables for Automated Mapping of Charcoal Kiln Sites. Archaeological Prospection, 22, 45-62. Stumpf, A., and Kerle, N. 2011. Object-oriented mapping of landslides using Random Forests. Remote Sensing of Environment, 115, 2564-2577. Zingman, I., Saupe, D., and Lambers, K. 2015. Detection of incomplete rectangular contours with application in archaeology. Technical Report, University of Konstanz. 28

    ×