Ship target detection method for SAR image of FNLM filtering combined with faster R-CNN

Abstract

Realizing the rapid detection of ship targets is of great significance both in commercial shipping activities and in modern warfare. Compared with visible light remote sensing images, synthetic aperture radar (SAR) has good resolution characteristics for metal targets, making ship targets more clearly visible in high-resolution SAR images. However, there are often obvious coherent bright spots and sea clutter noise in high-resolution SAR images used for ship detection, which seriously affects the detection accuracy. This paper proposed a detection method that uses fast non-local means (FNLM) filter and deep learning methods. Using high-resolution SAR images as the data source, FNLM and deep learning methods are used to realize the detection of ships quickly and accurately. First, the FNLM filter was used to preprocess highresolution SAR images to reduce overall image noise while enhancing target definition and feature details; Then, the Faster R-CNN algorithm was used to conduct model training on large-scale SAR data sets, to extract detection features, to improve the convergence accuracy of the network, and to achieve automatic and rapid detection of ships on the sea. Experimental results showed that this method had good robust characteristics and target detection accuracy, and the average accuracy of ship target detection reaches more than 90%. In the case of severe sea clutter interference, it still had a better detection effect.