Recognition Based On Neural Network For Motorial Object Above The Water

Intelligence and high speed are always important directions of ship research. As the speed of ship increasing, we need ship more intelligent. Vision is an important approach to acquire information. Therefore, the development of vision intelligence is important to enhance ship's independent ability, which makes ship recognize target exactly, response to corresponding information and complete the task successfully.Firstly, considering the characteristic of moving objects above the water, we do simulative experiments in the flume. Images of ship models in different circumstances and attitudes are acquired to build an image database.Secondly, after median filtering the images, on the basis of the real-time characteristic of moving objects, a feasible and effective method—Otsu is presented to segment images. Finally we acquired satisfactory effect of segmentation. The eigenvectors of images are got through the feature extraction and feature selection, which are the six feature vectors. Through 300 frames of four objects' images, we know that the six feature vectors had better clustering effect to congener objects and separable effect to different kinds of objects.Finally, we mainly do research on artificial neural network for identification. Considering the limitation of traditional BP neural network, we use a new Particle Swarm Optimizer method. After analyzing the algorithmic theory and parameter setting, we designed the program. Experimental results show that Particle Swarm Optimizer has a good astringency and can avoid the defect of BP neural network. Colligating Particle Swarm Optimizer and BP neural network, we can acquire satisfactory effect.According to the complex of the vision intelligence and information disposing, the paper is only the groundwork. However, as the available explore, methods mentioned in the paper lay the foundation for the further work.