Output details
15 - General Engineering
University of Sussex
Performance assessment of the modified-hybrid optical neural network filter
Presented is a new method that uses a non-linear neural network to train an image classification filter. This can then locate targets in difficult images that contain large amounts of clutter, noise and false targets. The non-linear mechanism is shown to outperform traditional state of the art linear correlation based classification filters and it can still be implemented as a high-speed optical computer. The work presents a rigorous analysis and is of interest to academics who work in object detection; and industry, especially defence companies who are trying to track and recognise specific targets within a video image.