The Research On Recognition Of Surface Targets Based On Neural Network

As people pay more and more attention to the ocean,it is becoming increasingly important to realize the classification and recognition of the surface targets quickly and effectively.For this problem,this paper takes the surface target as the research object and uses ImageNet data set as experiment data.The specific research of surface target recognition is carried out through the BP neural network and convolutional neural network.The whole content of this paper includes the following aspects.First of all,this paper introduces the relative technologies used in target recognition including the methods of image feature extraction,the bag-of-word,the parameter optimization in convolution neural network and the way to prevent overfitting.Secondly,we establish a relatively complete feature database of surface targets by extracting the 14 dimensional feature vectors,including the shape features,hu invariant moments and affine invariant moments.BP neural network is applied to train the feature vectors in the feature library.By adjusting the network parameters and learning steps,we can effectively classify and identify the surface targets and carry out simulation experiments under the Matlab 2014 a platform.Finally,we annotate several water target images in ImageNet data sets,and enhance the data by randomly flipping the images and changing contrast to get more training set and test set required by experiments.We construct the water target recognition network based on the AlexNet network.The experimental results show that the final recognition rate is over 90%,which can be effectively applied to the classification and recognition of surface targets.