Detection Of Ship Target In Optical Remote Sensing Image Based On Feature

More and more optical remote sensing images are available along with the advances in the data obtaining ability, it's urgent to improve the automatization level of optical remote sensing image interpretation. Ship target extraction from optical remote sensing image aims at extracting ship quickly and accurately from huge image, is the important subjects of optical remote sensing image interpretation.Combined with the practical demand in the correlative scientific research project that construct the ship targets detection subsystem from the optical remote sensing image, the optical remote sensing image preprocessing,ship target ROI extraction,ROI discrimination,ship team extraction is analyzed theoretically and validated by experiments.In the optical remote sensing image preprocessing aspect, studied to remove the image noise and separate the sea and land faced to the ship target extraction. Analyzed the method of selecting the filter window size in the noise removing using median filter. Studied to use an OSTU method combined with morphology processing latterly to separate the sea and land in the optical remote sensing image, validity is validated by experiments.In the target ROI extraction aspect, in view of the two cases that target pixel value is brighter and darker than the background, proposed two ROI quickly extraction method respectively. In the brighter case, drawing lessons from classic CFAR thought, proposed a quick CFAR detection algorithm based on gauss mixed model two level censoring. it considered the traits of the cloud and target, removed cloud and detected targets simultaneously, in the time of ensuring detection rate and false alarm rate, improved the efficiency obviously; in the darker case, studied a OSTU method based on the local variance, it could extract the target ROI quickly at the low operation cost.In the ROI discrimination aspect, mainly studied to extract the length and width of the ship target adopting the Minimum Enclosing Rectangle algorithm, and the ship wake feature adopting improved local Radon transformation. Also studied the sequential discrimination algorithm that combined target feature with ship team knowledge. Experiment show that it could remove the false ROI effectively.In the ship team extraction aspect, analyzed the spatial characteristic of the ship team, summarized the spatial regularity existed in the ship team which are collinear and equal spacing,parallel and equal spacing approximately. Proposed a ship team extraction algorithm using spectral graph partitioning based on spatial regularity. It established fuzzy regulation based on the spatial regularity, and measure the spatial regularity using the fuzzy logic technology, introduced the idea of spectral graph partitioning. It computed the eigenvector of the scene structure graph to partition the ship target, extract the ship team in this way. The experiment by emulational and real images show that it was effective in the case of ship team order change in certain limitation.