Ship Recognition Based On Convolutional Neural Network

With the rapid development of maritime trade,the importance of maritime security has become increasingly popular.The realization of automatic identification of maritime ship targets has a profound impact on both civil ship trade and military activities.In real life,the traditional ship automatic positioning technology and video monitoring system have the defects of low recognition accuracy,and the recognition effect needs to be improved.In image recognition,the existing intelligent recognition algorithms often have low recognition accuracy because of the poor meteorological conditions of the image to be recognized,the complexity of the shore based background and the small target of the ship to be recognized.In this paper,based on the research of traditional image features,the algorithm of yolov3 recognition based on convolutional neural network is improved to recognize ships under complex sea conditions.The main contents of this paper are as follows:1.Based on the analysis of deep learning and target detection algorithm,build the yorov3 network to classify the ships under the normal and complex sea conditions.It is found that the original network architecture has a good effect on ship recognition under the normal sea conditions,while the accuracy of ship recognition under the complex sea conditions still has a lot of room to improve.2.A hierarchical fusion algorithm based on yolov3 is proposed to improve the accuracy of ship identification in complex sea conditions.Yolov3 network consists of four parts:multi-scale training of samples,hierarchical feature extraction,selection and generation of ROI,and ship classification.The multi-scale training of samples can effectively increase the number of small target samples and balance the distribution of different types of ship samples,so that the ship recognition network can fully extract small target features.The convolution layer of the convolution neural network is used to extract the features of different layers of ship targets,and then the hierarchical feature fusion algorithm is used to combine the bottom features to form effective image recognition features.Through the non maximum suppression algorithm,the candidate boxes of interest are selected to determine the position of the candidate ships.The combined features are mapped to multiple groups for classification,and finally the maximum classification probability is selected to determine the candidate ship types.3.Through experimental verification,the proposed level fusion algorithm based on yolv3 is 10.91%more accurate than the original yolv3 deep learning target recognition algorithm,which can effectively solve the problem of ship recognition in complex sea conditions.