Recognising aircraft: automatic extraction of structure by layers of quadratic neural nets

This paper explores a method of recognition applicable to a wide range of problems. It adopts a syntactic or structural approach, which involves viewing an object in terms of its simpler components. In the specific example of aircraft recognition, this entails decomposing an aircraft into entities such as a nose, fuselage, tail and wings. Each component may then be further decomposed as a set of pixels. The process of recognition requires finding sets of pixels forming such components and then sets of these components forming an aircraft. This is implemented by a quadratic neural network