Multi-targets recognition for surface moving platform vision system based on combined features

The vision system of surface moving platform, such as military vessels, unmanned boat etc., is an important equipment for avoidance, target tracking and recognition. Farther offshore, an image captured by a camera usually includes water, air and targets; obvious surface targets generally include reef, islands and ships etc. Aiming at many research about surface target recognition focused on ships without thinking about the diversity of surface targets, this paper discussed mainly the feature extraction and recognition methods of multi-types targets. Firstly set up four kinds data source, including all kinds of reef, islands and ships, were obtained by a real yacht, searching network, a hand-making remote vessel and 3D ship models; secondly, texture features and shape features of above multi-targets were extracted, and extracted shape features included outer contour features, geometric features and moment invariant features; and then, features library of three types of surface targets was built; then, using principal component analysis (PCA) method optimized the training samples of BP neural network (BPNN); lastly, using grading BP neural network realized recognition of three types of surface targets. Experimental results show that proposal features extraction and recognition methods of multi-types targets based on combined features and PCA-BPNN can recognize effectively multi-types targets with a higher recognition rate over than 90%.