The Application Of Neural Network For Discriminating Littoral Oil-Spills By Laser Remote Sensing

Oil spills on the sea pollute the environment badly, so, it is very important to classify oil spills correctly in real time.In this paper , after comparing different classified methods of laser fluorosensors, an artificial neural network(ANN) approach, which is based on flexible nonlinear models for a very broad class of transfer functions, is applied for multi-spectral data analysis and differentiating between classes of oil on water surface. In this paper , two types of neural algorithm are investigated: Back-Propagation(BP) network and LVQ network . The reason of choosing BP and LVQ is that BP network is the common technique , it can approach any nonlinear functions with desired accuracy, but there is not widespread satisfaction with the effectiveness of this method since there is a danger of getting stuck or oscillating around a local minimum, so another algorithm ,that is LVQ network , is used in the paper. For both of them, a 3-layer neural network architecture with the input and single out nodes is used, the input is 64-channel spectra data, after training different neural network models based on spectra of samples, they generate the output of classification of unknown materials.In the test, the paper takes 30 groups of uncertain laser fluorescence samples as input .LVQ network is compared with BP network in the same condition. When the structure of BP network is 64-40-1, after it is trained 3259 times, the accuracy is 81.65%;when the structure of LVQ network is 64-70-8, after it is trained 373 times, the accuracy is 90.53%, and when the structure of LVQ network is 64-150-8, after it is trained 455 times, the accuracy is 92.30%.It shows that LVQ network as classifier for littoral oil-spill is better than BP network in both of the accuracy and velocity, and also LVQ is not easy to lead the system breakdown . So, it shows that LVQ network is a promising way to classify oil-spills on the sea.