The objective of the project is to develop image classifiers using neural networks based on the neurocomputer SYNAPSE1. This includes refinement of a graphical user interface called ECANSE for rapid prototyping.
Applications have been chosen from very different fields including magnetic resonance imaging, microcalcification detection in mammography and remote image processing. Use of the neurocomputer can speed up applications involving neural nets by up to three orders of magnitude. In particular research with large neural nets is feasible only with such specialised hardware.
The tasks of this project are:
- Development of the SYNAPSE Toolbox as part of the ECANSE graphical programming environment.
- Definition and labelling of representative medical and satellite images.
- Evaluation of different network types for applications.
- Development of a demonstrator for classifying microcalcifications in X-Ray images to find regions of interest more quickly.
- Development of a demonstrator for segmentation of MR Images of the brain.
- Implementation of neural network applications on the neurocomputer SYNAPSE1 to reduce the learning time by up to three orders of magnitude.