Research On The Detection Of Ship Targets On The Sea Surface In Optical Remote Sensing Image

Ship detection in optical remote sensing images is of great significance in both civil and military fields.In the civil field,it can be used for sea transportation,fishery management,assisting in marine rescue of ships in distress,etc.In the military field,it can be used to monitor the deployment and dynamics of ships in the key ports of the enemy and evaluate the effect of sea attack in wartime.However,due to the large size,long distance and complex imaging environment of remote sensing image,the abovementioned facts about VRS(visible remote sensing)images complicate the background and pose three main challenges to ship detection:· The high variability of targets caused by the viewpoint variation,imaging sensor parameters,occlusion,ship wakes,color,speed,and material of ships,etc.· High false alarm rate due to islands,heavy clouds,ocean waves,and the various and uncertain sea state conditions,like partial cloud cover,fog,wind,and swell.· The third issue is the computation burden.Most detection methods have high computational cost.Hence,reducing computational cost is considered to be a key issue for the large-scale remote sensing images.This paper designs a series of fast and effective ship detection algorithms in remote sensing image based on candidate region extraction,image segmentation,feature extraction and other technologies,using the detection mechanism from coarse to fine.The main research contents are as follows:1)Since the ships in a VRS image of the sea are salient objects,they are usually sparsely distributed and can easily be identified by the human visual attention system.Thus,the saliency models are introduced to identify attention-grabbing regions which may contain salient objects.Currently,most of the existing saliency algorithms are designed for natural images.When they are applied to remote sensing images,they have some drawbacks,such as the low resolution of saliency map,the low target integrity,and the blurring of the target boundary.In order to solve this issue,we construct a novel visual saliency detection method according to the difference of statistical characteristics between highly non-uniform regions which allude to regions of interest(ROIs)and homogeneous backgrounds.It can serve as a guide for locating candidate regions.Besides,it uses multi-scale features and multi-layer cellular automata fusion technology to solve the problem of variable target size.The algorithm can suppress the interference of cloud,sea clutter,island and other natural background to a certain extent.Moreover,it enhances the continuity of the whole target and the distinguishability between the targets,and extract more accurate candidate regions.This algorithm lays a good foundation for the following steps of feature extraction.2)Aiming at suppressing the distractors such as massive clouds,islands and sea clutter,the false alarm elimination method based on shape and texture features is designed.To solve the problem of target rotation,many methods rotate the target to a unified direction according to the pre-calculated main direction.However,if the main direction calculation is not accurate,the accuracy of the subsequent classification and recognition will be affected.To get a better representation of the target,both shape and texture features characterizing the ship target are extracted for subsequent classification.Moreover,the combined feature is invariant to the rotation.The shape descriptors based on global radial gradient transform and the texture descriptors based on regional covariance decomposition are concatenated as a feature vector,and then support vector machine is used to eliminate false alarms in candidate regions.3)In order to solve the problem of variable size of ship target,the steerable pyramid for multi-scale analysis is performed to generate a set of saliency maps in which the candidate region pixels are assigned to high salient values.And then the set of saliency maps is used for constructing the graph-based segmentation which can produce more accurate candidate regions compared with the threshold segmentation.More importantly,the proposed algorithm can produce a rather smaller set of candidates in comparison with the classical sliding window object detection paradigm or the other region proposal algorithms.Second,in the target identification phase,a rotationinvariant descriptor,which combines the HOG cells and the Fourier basis together,is investigated to distinguish between ships and non-ships.Meanwhile,the main direction of the ship can also be estimated in this phase.The overall algorithm can account for large variations in scale and rotation.Experiments on optical remote sensing(ORS)images demonstrate the effectiveness and robustness of our detection system.4)In order to detect the ship targets of small size more accurately,the YOLOv3 model is studied in depth,We improved the YOLOv3 model in the aspects of frame structure and loss function.The algorithm can effectively suppress the interference of complex background,and is more suitable for the detection of small targets.To sum up,this research work designs the algorithms suitable for ship target detection in optical remote sensing images,which provides a new idea in candidate region extraction and feature extraction.They are of reference significance for target recognition from satellite images.