Ocean Red Tide Recognizing Method Based Neural Network Ensembles

There are almost more than 4,000 sorts of algae which could result in the red tide in the world, but only two or three, named the dominant species, place a premium on red tide at a time. This paper presents a method which uses the hyper-spectral images of different familiar dominant species to train the different networks respectively, then synthesizes the outputs of the networks with the same weight to recognize the red tide. It not only conquers the difficulties that are the selection of the training data and the network´s training method, but also improves the generalization ability of the network system effectively. On the other hand, based on the neural network ensembles, the red tide recognition model could be extended easily and need not remodel the other networks. A mass of comparison experiments prove that the method recognizes the red tide and the dominant species effectively