1996 | OriginalPaper | Buchkapitel
High Resolution Image Reconstruction of Polymer Composite Materials Using Neural Networks
verfasst von : A. C. Pardoe, D. A. Hutchins, J. T. Mottram, E. L. Hines
Erschienen in: Review of Progress in Quantitative Nondestructive Evaluation
Verlag: Springer US
Enthalten in: Professional Book Archive
Aktivieren Sie unsere intelligente Suche, um passende Fachinhalte oder Patente zu finden.
Wählen Sie Textabschnitte aus um mit Künstlicher Intelligenz passenden Patente zu finden. powered by
Markieren Sie Textabschnitte, um KI-gestützt weitere passende Inhalte zu finden. powered by
A neural network is an artificial intelligence technique inspired by a simplistic model of biological neurons and their connectivity. A neural network has the ability to learn an input-output function without a priori knowledge of the relationship between them. Typically a neural network consists of layers of neurons, whereby each neuron in a given layer is fully connected to neurons in adjacent layers. Figure 1 shows such an arrangement with three layers, called the input, hidden and output layers. The connection strengths between neurons, often referred to as weights, are modified by a training phase. The training phase used here utilizes an error back propagation algorithm [1]. During training the neural network is presented with input which propagates through the network producing a corresponding output. A comparison of the actual output with the desired or target output generates an error which is used to adjust the neural network’s weights according to an error gradient descent technique [2]. This procedure is repeated for many different input and desired output pairs allowing the neural network to learn the input-output function.