Abstract
It is still challenging to effectively detect ship objects in optical remote-sensing images with complex backgrounds. Many current CNN-based one-stage and two-stage detection methods usually first predefine a series of anchors with various scales, aspect ratios and angles, and then the detection results can be outputted by performing once or twice classification and bounding box regression for predefined anchors. However, most of the defined anchors have relatively low accuracy, and are useless for the following classification and regression. In addition, the preset anchors are not robust to produce good performance for other different detection datasets. To avoid the above problems, in this paper we design a paired semantic segmentation network to generate more accurate rotated anchors with smaller numbers. Specifically, the paired segmentation network predicts four parts (i.e., top-left, bottom-right, top-right, and bottom-left parts) of ships. By combining paired top-left and bottom-right parts (or top-right and bottom-left parts), we can take the minimum bounding box of these two parts as the rotated anchor. This way can be more robust to different ship datasets, and the generated anchors are more accurate and have fewer numbers. Furthermore, to effectively use fine-scale detail information and coarse-scale semantic information, we use the magnified convolutional features to classify and regress the generated rotated anchors. Meanwhile, the horizontal minimum bounding box of the rotated anchor is also used to combine more context information. We compare the proposed algorithm with state-of-the-art object-detection methods for natural images and ship-detection methods, and demonstrate the superiority of our method.
Highlights
Ship detection in high-resolution optical remote-sensing images has been widely used in military and civilian fields
Compared with horizontal anchor box, extracting features for rotated anchor box is more complicated [51]. To both reduce computation and use context information, in this paper, we propose to take advantage of horizontal minimum bounding box of rotated anchor box instead of using the enlarged anchor box
The proposed detection method is compared with three other detection algorithms, including Faster R-convolutional neural networks (CNN) [9], SSD [10], and Rotation Dense Feature Pyramid Network (R-DFPN) [16]
Summary
Ship detection in high-resolution optical remote-sensing images has been widely used in military and civilian fields. Rotated bounding boxes are predicted from rotated anchors for one-stage detectors [14,17], and from rotated region proposals for two-stage detectors [16] Both object detectors for natural images and ship detectors for remote-sensing images perform accurate object (ship) detection by predefining some default anchors. In order to robustly detect objects with various sizes and shapes, multiple scales and aspect ratios (and angles for ship detectors) are usually needed to be set, so that the defined anchors are able to adapt as many objects in real scenarios as possible. In this way, many anchors would be generated. Experimental results demonstrate that the proposed method can accurately locate ships in remote-sensing images with complex backgrounds
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have