Feature Extraction And Classification Of Ship Targets In SAR Imagery

As an active imaging system,Synthetic Aperture Radar(SAR)is not limited by the external environment such as the weather,light and clouds,and can observe the ocean continuously.The recognition of SAR ship target is one of the important contents of SAR image interpretation.Its purpose is to distinguish the category information of each sample.With the development of SAR technology,the recognition of SAR ship target has received more and more attention,and has become an important research direction in the field of remote sensing.Focusing on the recognition of SAR image ship targets,this paper studies the topic from two aspects,i.e,the traditional machine learning methods and current rapidly developing deep learning methods.Focusing on the recognition of SAR image ship targets,this paper mainly studies the extraction of ship target features,the large-scale,medium-scale and small-scale ship target classification based on the tradictional classifier SVM,and the deep-learning based convolutional neural network,as well as the classification of different types of ship targets based on the convolutional neural network combined with the spatial pyramid pooling and the metric learning.The main contents are summarized as follows:1.The feature extraction of SAR image ship targets is studied,and the large-scale,medium-scale and small-scale ship target classification is completed in combination with the traditional classifiers.First,the pre-processing and feature extraction operations are carried out on the ship targets;and then some prior knowledge is used to discriminate the rationality of the extracted features;then the preliminary results of the pre-processing and feature extraction are corrected to obtain the final result.Secondly,the classification of large,medium and small ships based on SVM is studied.In addition,the classification of ship targets based on ensemble learning is also studied.At first,the bagging method in ensemble learning is introduced.Then,the bagging in ensemble learning combined with SVM is applied to the classification of large,medium and small ships.And the experimental results show that,compared with the classification based on SVM method,the bagging method in ensemble learning combined with SVM obtains better classification results.2.The classification of large-scale,medium-scale and small-scale ships based on convolutional neural networks is studied.Firstly,the basic principles of artificial neural network and convolutional neural network are introduced in details.The convolutional neural network suitable for the data used in this paper is designed.The influence of various hyperparameters of the convolutional neural network on the classification performance and the convergence speed is studied.The classification results of the traditional machine learning method are compared with the classification results of the deep-learning based method.The experimental results show that the classification of large-scale,medium-scale and small-scale ships based on convolutional neural networks can extract more effective feature and obtain higher classification rate.3.The two learning strategies,spatial pyramid pooling and metric learning,are combined with convolutional neural networks to classify different types of ship targets.Firstly,the classification of different types of ship targets based on spatial pyramid pooling combined with convolutional neural network is studied.The difference from convolutional neural network is that the last layer of the pooling layer is replaced by the spatial pyramid pooling,and the multi-scale information can be extracted from images.Secondly,the metric learning is briefly described,and the distance metric of the triplet samples is added to the loss function of the convolutional neural network.The above two schemes were verified by using measured SAR data.From the experimental results,it is found that the above two algorithms are superior to the method using the convolutional neural network for the classification of different types of ship targets.