| Ship surveillance is one of the most important applications of synthetic aperture radar (SAR) imagery, as it can help to improve the activities of fishing control, marine traffic monitoring, exploitation of ocean resources, and marine environment protection. In recent years, there is rapid development of space remote sensing in China, which provides great data to support ship monitoring, as well as brings additional requirement for efficiency and precision. Therefore, the research on ship monitoring in SAR images is an advanced issue with great theoretical and applicable significances.This dissertation mainly focuses on several of the key techniques to improve the ship monitoring performance, including the ship detection in SAR images, feature extraction and classification of ships in high resolution SAR images. The main work and innovation of this dissertation are as follows:(1) To improve ship detection accuracy and reduce runtime, the dissertation proposes a ship detection method based on Deep Neural Network (DNN). First, we choose unsupervised DNN as network architecture. Then, the subbands of SAR images after wavelet transform are used to train different DNNs. Finally, the outputs of DNNs are applied to features fusion and ship detection. The method is validated on TerraSAR-X images. The detection results demonstrate that the proposed method superiors to other detection method.(2) To improve ship classification accuracy, the dissertation proposes a ship classification method based on superstructure scattering features in SAR images. First, we analyze the scattering features of superstructure. Then, we propose a novel ship segmentation method based on peak extraction to reduce noise interference. Finally, a feature named ratio of dimension is proposed to describe superstructure. Ratio of dimension is adopted to classify ships via SVM. The method is validated on RadarSat-2 images. The detection results demonstrate that the proposed method superiors to other classification method. |
PDF Full Text Request