G1.6 the Application of Neural Networks to Image Segmentation and Way- Point Identification

The analysis of images taken of the ground from aircraft and satellites is of intense importance. This chapter describes work using the ADAM neural network that was aimed at finding way-points, roads, towns and rural areas in Infra Red Line Scan images of the ground. The network is capable of on-line, single pass training and simple computer implementation, making it particularly applicable in image processing tasks. In addition, no image pre-processing is needed, and images may be very large which makes the approach particularly simple to adopt. G1.6.1 Project overview. This project was aimed at finding way-points and finding roads, towns and rural areas (segmentation) in infra-red line scan (IRLS) images of the ground taken from aircraft. The problem of finding way points is vital for airborne vehicles (i.e. land marks that are used to guide a vehicle), and can be useful in many other image processing tasks. In this work we were particularly interested in a system that could be trained rapidly on new features and way-points. The more conventional network architectures, based on gradient descent learning could not be used in the problem because they could not be rapidly trained on the large and complex data sets used in the work. The use of the ADAM memory (Austin 1987) combined the fast feature recognition ability of the network with the large capacity of the network so that a large number of way points be stored.