2016 Fourth International Workshop on Earth Observation and Remote Sensing Applications 978-1-5090-1479-8/16/$31.00 ©2016 IEEE Fast Ship Detection from Optical Satellite Images Based On Ship Distribution Probability Analysis Xie Xiaoyang, Xu Qizhi School of Computer Science and Engineering Beihang University Beijing 100191, China elizabeth.xie@buaa.edu.cn; qizhi@buaa.edu.cn Hu Lei School of Computer Information Engineering Jiangxi Normal University Nanchang 330022, China hulei@cse.buaa.edu.cn Abstract—Automatic ship detection from optical satellite images remains a tough task. In this paper, a novel method of ship detection from optical satellites is proposed by analyzing the ship distribution probability. First, an anomaly detection model is constructed by the sea cluster histogram model; then, the ship distribution based on the ship safety navigational criterion is analyzed to obtain the ship candidates, and obvious non-ship objects are removed by the area properties from ship candidates; finally, a structural continuity descriptor is designed to remove false alarms from the ship candidates. Experiments on numerous satellite images from panchromatic and one band within multispectral sensors are conducted. The results verified that the proposed method outperforms existing methods in both effectiveness and efficiency. Keywords—Ship detection, optical satellite images, remote sensing I. INTRODUCTION Widespread commercial satellites lead to easier access in satellite images, hence the number of applications on satellite- images have increased. Ship detection applications based on satellite image have attracted great interests since they perform vital importance on maritime surveillance and national defense. In exiting works, optical images are considered more suitable than other remote sensing images for object detection because of its visualized contents [1]. With the enhancement of optical satellite image quality, richer information of ship and sea state can be acquired meanwhile increase the difficulty of object detection tasks. Hitherto, there are following challenges in ship detection from optical satellite images (SDOSI): (i) As further improvement in image spatial resolution, massive image data from optical satellite sensors lead to greater burden for real- time ship surveillance; (ii)methods must adapt to variable ship shape features due to different shooting angles and occlusion from shadows of hull structures; (iii)methods must resist to richer edge information within image from clouds, ocean waves and small islands which result in heavy false ship candidates and affects the performance of ship detection. Previous works have established a fine basic ship detection flow including three main stages: 1) sea segmentation; 2) ship candidates extraction and 3) ship candidates classification. The first stage is sea region extraction; the second stage is to extract areas that probably contain ship targets from sea region; the last stage is ship and non-ship targets classification in suspected target areas. Zhu utilized edge information in the second stage and local multiple patterns to discern ship targets in [2]. Tang [3] extracted ship edge and shape feature exploited deep neural network under low and high frequency subbands from JPEG2000 compressed domain. Each of these works relies heavily on the edge information of ship and may have poor performances when the sea is rough because of raising false alarm rate caused by edge features of ship wakes and ocean waves. On the other hand, Shi [4] employed contextual information constructed by spatially adjacent pixels to augment the separability between ship-targets and their background to obtain ship candidates. Yang [5] analyzed the local intensity and texture similarity of sea surface and proposed intensity discrimination degree to perform SDOSI. Each of these techniques exploited statistical features of grayscale between ship and the sea. However, they may not fully address the aforementioned first issue. Because satellites have limited resources for computation and calculation, real-time ship detection system’s performance and complexity should be considered as equally important. Unlike previous works which usually employed only features that targets presented in the images, proposed method combined ship features in image with ship behavior characteristics in the sea to perform ship detection. Practically, ship navigational criterion which stipulates safety navigational distance among vessels can be implemented with imagery features as ships generally obey it when sailing on sea. This paper mainly concentrates on the last two stages of ship detection considering sea region has been marked by prior geographic information. Additionally, only ship detection in sea and not in harbor on panchromatic band or a band of multispectral are interested in this paper. The contributions of this paper are threefold: (1) a new anomaly detection model is developed for fast ship detection; (2) the proposed ship density function is adapted to variant conditions; and (3) a novel structural continuity descriptor is adopted for false alarm removing, and thus, a real-time accurate ship detection system is achieved. This work was supported in part by the National Natural Science Foundation of China Key Project under Grant 61331017 and in part by the National High-tech R&D Program (863 Program) of China under Grant 2015XXXX73.