Research On Ship Detection From High-resolution Optical Remote Sensing Images

Target detection in remote sensing images has been a hot research topic in the field of target recognition.Ship target in remote sensing images has a very important strategic significance in both civil and military fields.High-resolution remote sensing images provide more detailed information which provides a new research direction on ship targets detection,but the time costs for processing such massive data brought by high-resolution images influences the detection efficiency.Therefore,efficiently and accurately extracting ships from high-resolution remote sensing images is a hot spot in object extraction from remote sensing images.This paper proposes a new method for ship detection based on machine learning.It is able to improve the detection accuracy and time efficiency from high-resolution optical remote sensing images.Deep learning is widely used in the field of target de-tection and recognition recent years,we introduce Faster R-CNN method into remote sensing image ship detection in this paper.We build up the CNN model suiting for ship target and compare the method of combining traditional machine learning with feature and the other method of ship detection by experiments.The result shows that method based on Faster R-CNN not only improves the detection accuracy,but also enhances the detection efficiency greatly.The main work of this paper includes the following parts.First of all,analyzing the basic idea of regulation-based traditional ship detection,and proposed a method for fast extraction of coastline.The edge detection and closed operation are used to detect the remote sensing image quickly,and the target infor-mation is extracted from target enhancement and the two value of the target.Secondly,in viewing the poor detection performance,low accuracy and weak generalization ability in regulation-based detection method,this paper proposes a kind of optical remote sensing image ship detection method combined with linear support vector based on HOG features.In sample acquisition stage,the ship samples are ro-tating to the same direction.Then,with the help of super-pixel segmentation and clustering algorithm,the negative samples of different scene categories are selected.In the ship detection stage,the target direction and the training sample are rotated by edge detection and Hough transform.Finally,the trained model is used to detect the ship targets in the image.Finally,according to the problems of the traditional machine learning method with complex preprocessing stage and the need to manually select feature,we intro-duce the current popular method of deep learning to perform ship detection,and pro-pose detection method based on Faster R-CNN and the corresponding parameters are adjusted to make the model suitable for the detection target.Enriching the sample di-versity by multidirectional rotation.At last,the proposed method is compared with other ship detection algorithms.The experimental results show that the proposed method can effectively solve the problem.