Study On Methods Of Ship Target Detection Based On Pol-SAR

As one of the hottest developments in global synthetic ape rture radar(SAR) systems, Polarization SAR(Polarimetric SAR, Pol-SAR) has full polarization measurement capability, and can work under different polarization combinations between receiver and transceiver. Compared with conventional single polarization SAR, polarimetric SAR can acquire more information. China has a very broad area of territorial waters. With the increasing maritime rights disputes, ocean monitoring ability needs to be improved. As a key marine monitoring, ship target detection is of great significance in safeguarding maritime rights, enhancing coastal warning, monitoring maritime traffic and other aspects. Therefore, ship detection methods for polarimetric SAR data are very necessary and meaningful. Moreover, the Chinese armed police force(CAPF) are perennially servicing for the customs inspection and anti-smuggling. With the sharp increase of the cases of smuggling and illegal immigration, how to effectively detect and monitor the illegal vessels is becoming a urgent problem to be solved for the CAPF. Accordingly, it is an urgent and practically-required need to study the vessel detection based on the polarimetric synthetic aperture radar data. The dissertation focuses on two approaches which based on reflection symmetry metric and polarization decomposition, respectively. The main works is as follows:1. Statistical modeling and parameter estimates of reflection symmetry metric were investigated. In order to achieve ship target detection, CFAR detection method can also be used. In previous literature, the threshold of reflection symmetry metric was always selected according to the experience when detecting targets. However, when the polarimetric SAR imaging system and sea background conditions change, the threshold will also change, which results in that the method has bad adaptability and is difficult in practical applications. Based on the theory of complex Gaussian multivariate statistical distribution, we use the complex Wishart distribution and combine the product model to analytically derive the statistic distribution of reflection symmetry detection. The proposed method takes advantages of the logarithmic cumulants to estimate parameters. The measured experimental data proves that derivation distribution can better fit the detection distribution histogram. Comparing with conventional CFAR ship detection method based on amplitude information, the proposed method has better performance.2. We select the volume, secondary and surface scattering component acquired by Freeman decompo sition, and combine polarization channel component as the feature vector of support vector machine. Then, we train the sample data to obtain a classifier that can be used to detect ships. Volume, surface and secondary scattering components(denoted by P_v, P_d and P_s respectively) based on Freeman decomposition, as well as polarization channel components are then used to construct a feature vector of a support vector machine(Support Vector Machine, SVM). On the basis of validating the effectiveness of the single feature value of the feature vector, this dissertation selects the samples form the real polarization SAR data and trains the SVM. We detect the vessel targets for the test data with the SVM models. The proposed method can retrain the false alarms with better performance than the CFAR method based on the reflection symmetric metric.