A novel ship detection algorithm based on anomaly detection theory for SAR images

摘要

Ship detection is a key topic in remote sensing. In existing research works, SAR images are part of the data sources which have been most widely studied because they can be obtained day and night, regardless of the impact from meteorological conditions. This paper mainly aims at the existing problems of intelligent detection algorithm for ship targets in SAR image, such as the uncertainty of the automatic censoring. A novel algorithm for ship detection based on anomaly detection theory was presented. Firstly, we transform the SAR image into hyper-spectral form by rearranging the spatially adjacent pixels into a vector, which combines the gray feature and context feature of pixels and improves the separation degree between background and ship, including the accuracy of separation. Then a hyper-spectral algorithm is applied to extract ship candidates preliminarily and quickly by regarding ships as anomalies. Finally, we choose a CFAR detector to extract ships more accurately. In the background clutter modeling stage, the clutter environment can be determined adaptively to prescreen the clutter pixels in the sliding window used for detecting. In the detection stage, we just need to examine the ship candidate pixels. The effectiveness of the algorithm is tested by the measured SAR images. Experimental results show that the algorithm has a better detection performance and high fidelity of target structure. Meanwhile, this algorithm can also reduce the number of false alarm and side-lobe.