The Research On Sea-ice Detection Method In Airborne SAR Images

The oceans account for 71 percent of the earth’s surface. It has become an important and indispensable part of human survival and development. The oceans have rich resources, and meanwhile maritime transport has become an important means of international logistics. But sea ice disaster has an enormous impact to human’s maritime transport and marine resources development. So, carrying out the research on forecasting the sea ice has become an important project for nationals and national defense construction.(Synthetic aperture radar) SAR is a new kind of radar, which uses the Doppler shift of relative motion between the radar and target to improve the resolution throughout the day. It can monitoring throughout the day and all-weather and can get high resolution SAR images. With the development of synthetic aperture radar, SAR uses widely into military and civilian areas. And now some departments gradually applied SAR on detecting the target in sea, forecasting marine disaster and monitoring sea conditions.This paper firstly pretreatment the SAR sea oil images, reduce the impact of speckle noise to image processing. The result shows that SRAD filter approaching image, as Frost filter and Lee filter, retain the characteristics of the target image effectively, and more reservations on the edge of image target.Before training with Support Vector Machine (SVM), we must first determine the input feature vector. According to the characteristics of gray level co-occurrence matrix (GLCM), this paper select five kinds of feature vectors, they’re energy, contrast, inverse difference moment, entropy and correlation. Besides, according to Canny operator algorithm, this paper also make acquired edge length and six kinds of edge direction as seven kinds of feature vectors. Finally compositing a group of 12-dimensional feature vectors as the training feature vectors.In this paper, firstly creating a SAR image database, take the 64*64 pixel sea ice and sea texture image. Using nuclear function of Mercer theorem, do multi-dimensional SVM regression training to get the texture features of sea ice, sea water and ice water mixing zone, and then do regression experiments. The results show that support vector machine method can effectively distinguish sea ice and sea of SAR sea ice image. And compared to similar training machine, it has better data and theoretical support, and the development prospect is huge.