Ship Detection Method Based on Gabor Filter and Fast RCNN Model in Satellite Images of Sea

Abstract

In this paper, we design a new approach which combines the Gabor filter with the principle of selective search algorithm applied in Fast RCNN to extract and enhance the texture of the satellite image, to solve the problems of low detection precision, high probability of missed detection and error detection of ships in satellite images. Experiments show that the texture enhancement method based on Gabor filter fits the principle of selective search algorithm, which increases the number of positive samples in region proposal. The mAP of ship detection in satellite images reached 75.21%. And the detection accuracy of moving ships increased by 10%.