Classification of wood species by neural network analysis of ultrasonic signals

Abstract The passage of ultrasonic waves through an anisotropic inhomogeneous material such as wood involves complex interactions between the physical vibrations of the ultrasound and the elastic response of the wood. The initial ultrasound signal is modified by the transmission medium in a way characteristic of the elastic anisotropy of the medium. The many species of wood have subtly different elastic responses. In this work the characteristic signals formed by these responses is examined. A neural network system is used to classify these signals in terms of species. The neural network is shown to have a high success rate in identifying wood species from the ultrasonic trace. It is established that this identification is not possible using wave velocity or received signal amplitudes. The most appropriate propagation direction for species identification is also considered.