A Neural Network Approach To Underwater Target Tracking
Ling Guan · 2005
Abstr<act - An underwater target tracker based on neural network structure is introduced. The neural network is of feed-forward or recurrent type. Instead of using the standard training which requires the unrealistic ideal data, it is trained by a classical filtering criterion. The neural tracker has proved to possess some unique propertiies: 1. The performance of the tracker is similar to its; teacher, but better than conventional trackers which use less sophisticated criteria.; 2. It is very efficient in real-time tracking since only si3mple mathematics is involved, the computationally intensive training is done off-line; 3. 13ecause of its efficiency, many computationally intensive tracking methods can be realized. The neural tracker is tested by synthetic examples. The results show that while the performance is as expected, a significant saving in time is achieved.