Research On Target Recognition Algorithm Of Maritime Ship

The ocean contains abundant resources,and sea transportation has many advantages over other modes of transportation.With the emergence of unmanned ships at sea in recent years,the traditional manual maneuvering of ships is changing.Intelligence will be the development direction of future ships.In order to ensure the safety of unmanned ships,ships need to be able to accurately sense the surrounding environment,and use a variety of sensors and artificial intelligence technology to establish a target recognition system for maritime ships,which can provide security means for maritime navigation of ships.This article attempts to apply SSD(Single Shot Multi Box Detector)network to the research of maritime target recognition algorithms to identify the category of ships at sea.First of all,by analyzing the shape characteristics of various maritime ship targets,the maritime target classification method applicable to this paper is summarized.Using the original samples obtained by the web crawler,a ship sample database for maritime target recognition was established in this paper.In the process of establishing the sample database,a data augmentation method was applied to realize the expansion of the quantity and quality of the ship sample database.The sample data is more diverse,which improves the generalization ability and robustness of the network model.Secondly,this paper compares the three most commonly used convolutional neural network models,Alex Net,Vgg Net-16,and Goog Le Net,and two mainstream deep learning frameworks,Caffe and Tensor Flow,and uses experiments to compare and analyze the performance of the three networks.Then the algorithm of marine ship target recognition based on SSD network is studied in detail.Experiments show that the algorithm has a fast training speed and high recognition accuracy,and it can complete the recognition task well under difficult conditions such as occlusion and blurred target recognition,and it is easy to improve the network structure under the Caffe framework.Aiming at the problem of losing some valid information in the feature extraction process,this paper uses a multi-level feature fusion convolutional neural network.The specific operation is to fuse the feature maps of the 4-3 convolution layer and the 5-3 convolution layer in the SSD basic network Vgg Net-16,and retain more valid information of the same image into the classifier at the same time.Through experimental comparison,the recognition accuracy of the multi-feature fusion technique used in this paper is significantly better than that of ordinary convolutional neural networks,and is much better than traditional neural network recognition algorithms.The thesis gives the establishment of the above neural network model and algorithm training method.The recognition of a series of marine ship targets verifies the stability and effectiveness of the proposed method.Finally,the work done is summarized,and the deficiencies and possible future research directions of the sea ship target recognition algorithm proposed in this paper are evaluated.