Ship Detection In Optical Satellite Image Based On PCAnet

The ship detection is widely demanded in both civil and military fields,so the research on ship detection has very important significance to the sea surveillance of our country.In recent years,with the rapid development of optical imaging technology,the optical satellite image with its outstanding advantage in the ship detection area has drawn the highest attention of the marine monitoring department and many scholars,and a growing number of the relevant research literature have been published.The thesis has made some study focusing on the ship detection in optical satellite image,and the major work includes the denoising preprocessing of the optical satellite image,the predetection of ship target and the ship target identification.1.As to the denoising preprocessing of the optical satellite image,the ranked-order-based adaptive median filter(RAMF)is adopted in the thesis to remove the noise in the optical satellite image,and the classical wiener filter is used to remove gaussian noise in the optical satellite image.Experimental results show that the denoising method adopted in the thesis can not only effectively remove the noises but also retain the details of ship targets in the optical satellite image.2.In terms of the ship target pre-detection,a ship target pre-detection method based on the RX(Reed-Xiaoli)anomaly detection algorithm is applied in the thesis.Since ships can be regarded as anomalies relative to the vast sea background,the ship candidates in the optical satellite image can be regarded as anomalies and can be preliminarily detected by the RX algorithm.Specifically,this predetection method first transforms the original optical satellite image into a three-dimensional cube image,then inputs the three-dimensional cube image to the RX algorithm to preliminarily detect ship candidates in the optical satellite image.3.As for the ship target identification,we propose a ship target identification method which combines a deep learning network:principal component analysis network(PCAnet)with the support vector machine(SVM)in the thesis.This ship target identification method first utilizes the PCAnet to extract features from samples and ship candidates,and uses the features of samples to train the SVM classifier,then utilizes the SVM classifier to achieve binary classification for the features of ship candidates.Finally,the ship target identification method identifies real ships among ship candidates and removes false alarms.Experimental results show that the proposed ship target identification method can effectively improve the detection accuracy,and can reduce the false alarms resulted from interference factors such as the waves and clouds in the optical satellite image to reduce the false-alarm rate.