S-CNN-Based Ship Detection From High-Resolution Remote Sensing Images

Abstract

ABSTRACT:Reliable ship detection plays an important role in both military and civil fields. However, it makes the task difficult with high-resolution remote sensing images with complex background and various types of ships with different poses, shapes and scales. Related works mostly used gray and shape features to detect ships, which obtain results with poor robustness and efficiency. To detect ships more automatically and robustly, we propose a novel ship detection method based on the convolutional neural networks (CNNs), called S-CNN, fed with specifically designed proposals extracted from the ship model combined with an improved saliency detection method. Firstly we creatively propose two ship models, the "V" ship head model and the "||" ship body one, to localize the ship proposals from the line segments extracted from a test image. Next, for offshore ships with relatively small sizes, which cannot be efficiently picked out by the ship models due to the lack of reliable line segments, we propose an improved saliency detection method to find these proposals. Therefore, these two kinds of ship proposals are fed to the trained CNN for robust and efficient detection. Experimental results on a large amount of representative remote sensing images with different kinds of ships with varied poses, shapes and scales demonstrate the efficiency and robustness of our proposed S-CNN-Based ship detector.

Vessel detection and classification from spaceborne optical images: A literature survey

“…A relatively large fraction of the authors (35%) used Google Earth images as the direct input data source in their research on vessel detection – either as exported data or simply as a print screen –, or as a source for collecting the greater amount of test data for machine learning methods (An et al, 2013, p. 201; Deng et al, 2013, Dong et al, 2013, Gan et al, 2015, Guo et al, 2015, Han et al, 2014, Hong et al, 2007, Huang et al, 2016, Johansson, 2011, Ju, 2015, Ma et al, 2010, p. 201; Shi et al, 2014, Xu et al, 2017, Xu et al, 2011, p. 201; Xu and Liu, 2016, Xu et al, 2014, Yang et al, 2017, Yang et al, 2014, You and Li, 2011, Zhang et al, 2016, Zou and Shi, 2016). One author has used data from Microsoft Virtual Earth (Yin et al, 2016).…”

“…Wang et al2016Orbview-3PAN, B, G, R14n/aAnomaly detection method: Probability density function, vessel distribution by the densityDiscrimination: structural continuity descriptor (based on width to length ratio)yesVessel detection on open seaXiaoyang et al2016Google EarthPAN−−n/aSalient-based method: hypercomplex frequency domain and phase quaternion Fourier transformDiscrimination: radon transform and histogram of oriented gradientsyesVessel detectionXu and Liu2016n/aPAN2−40.0Computer vision method: AdaBoost classifier trained by Haar features, Line Segment Detector−yesVessel detectionYao et al2016Microsoft Virtual Earthn/a−4n/aSalient-based method: phase spectrum of Fourier transform saliency and frequency-tuned saliencyClassification: based on geometric characteristics, SVM classifieryesVessel detectionYin et al2016Google Earthn/a0.5, 1−n/aSalient-based method: normal directional lifting wavelet transform−yesTarget detectionL. Zhang et al2016Google Earthn/a0.12, 0.25−n/aDeep learning method: ship proposal extraction convolution neural networks−yesVessel detectionR. Zhang et al2016GaoFen-1, VRRS-1, Google EarthPAN2 (GF-1), 16 (VRSS-1)−20 pixelsDeep learning method: convolutional neural network, Singular value decomposition algorithmDiscrimination: SVM classifieryesVessel detectionZou and Shi2017Google Earthn/a−−n/aComputer vision method: rotation and scale-invariant method based on the pose consistency voting−yesInshore vessel detectionHe et al2017Google EarthPAN2−…”

A Coarse-to-Fine Network for Ship Detection in Optical Remote Sensing Images

“…These methods intuitively extract features of images through CNN, avoiding complex shape and texture analysis, which significantly improve the detection accuracy and efficiency of ships in optical remote sensing images. Zhang et al [38] proposed S-CNN, which combines CNN with the designed proposals extracted from two ship models. Zou et al [23] proposed the SVD Networks, which use CNN to adaptively learn the features of the image and adopt feature pooling operation and the linear SVM classifier to determine the position of the ship.…”

An Adaptive Sampling Scheme to Efficiently Train Fully Convolutional Networks for Semantic Segmentation

“…Other applications, where training CNNs on very large images is a problem, includes the segmentation of histology datasets [5] or the segmentation of aerial images. For example in aerial image segmentation, training a CNN to segment ships [6] can be difficult because large portions of the image contain water which provide little information during training, resulting in slow learning. Some ideas to address this have already been proposed, for example in [3] a fixed, handcrafted, pre-computed weight map is used to help learn small separation borders between touching cells for biomedical image segmentation.…”