Neural networks are certainly a very promising technique for target recognition, because of their adaptability, their fault tolerance and their real-time potential due to their parallelism. If, as in most of the applications of neural networks, database availability, choice of preprocessing, and features extraction are important to keep the amount of time necessary for learning within reasonable limits, the key factors for the success of the applications are multi resolution recognition capabilities, invariance of recognition by translation, rotation, scaling, movement detection capability. The integration of neural modules in weapon systems requires new validation processes, as well as a careful study to make the neural modules compatible with the sequence of functionalities of the system. Backpropagation is certainly the most often used neural algorithm, because of its ability of extracting features. Various comparisons of performance with classical methods have been made on some examples. One is given, where neural networks outperform classical algorithms for some problems of classification of natural underwater sounds. But there is no general rule, and in fact, most of the time performances mainly depend on the representativity of the database.
Neural networks for target recognition
Neuronale Netze für die Zielerkennung
1991
12 Seiten, 21 Quellen
Aufsatz (Konferenz)
Englisch
Neural networks for automatic target recognition
Tema Archiv | 1990
|Target recognition and classification using neural networks
Tema Archiv | 2002
|Evaluation of Neural Networks for Automatic Target Recognition
British Library Conference Proceedings | 1997
|Target recognition using neural networks for model deformation measurements
Tema Archiv | 1999
|