Ship Target Detection Technology In Remote Sensing Image

As an important means of transportation in the vast ocean,ships have always been the key objects of ocean target detection in remote sensing images.The purpose of ship target detection in remote sensing images is to obtain the interested ship targets from a large number of image data,and obtain its position,course and other information in time,which is the premise of ship targets tracking and recognition.There may be both land and ocean areas in remote sensing images.The accuracy of ship targets detection would be affected due to the complex environment of land such as buildings,shadows,trees and roads.Therefore,it is necessary to divide land and sea before detection to avoid interference from land areas as much as possible.In this paper,sea-land segmentation is performed by threshold segmentation and regional growth.Morphologica l method is used to make up for the over-segmentation of threshold method,and then the coastline is accurately extracted through regional growth.The grayscale and texture characteristics of shore ships are close to those of land,so they are usually divided into land.In this paper,based on its docking characteristics,the area near the coastline is selected as the ROI(region of interest)for the detection of inshore ships.In this way,most of the interference from the complex background could be well removed.Since the bow of the ship has an obvious V-shaped feature,in this paper,we first detect the SUSAN corner point according to this feature,and extract the square area to detect the straight line.Then,the bow is confirmed and the direction of the ship is calculated according to the result of the line detection.Finally,the body of the ship is detected along its direction,and ship target is extracted.Experimental results have shown that the recall rate of this method is 79.23% in different environments.For ships with V-shaped features,it has a good detection effect,but straight lines and corners are difficult to detect on ships that are not V-shaped at the head.To solve this problem,we design a detection method based on bow contour matching and body detection in this paper.The method is based on Hausdorff distance for contour matching.Since Hausdorff does not have the invariance of rotation and scale,we first convert the image to the log-polar coordinate system to translate the rotation and scaling into translation along the distance and angle axis.Then bow matching is carried out in the log-polar coordinate system to extract the bow region.After that,the ship direction is calculated by the phase correlation method.Finally the false alarm is removed by information of body detection and the detection results are obtained.Experimental results have shown that the results of our method is better than other algorithms in the literature,and the detection accuracy is 92.86%.Aiming at solving the uneven light and shade phenomenon of remote sensing images and the interference of complex background such as islands and clouds,a detection method for offshore ships based on saliency detection and feature extraction is designed in this paper.Firstly,the input image is processed by adaptive Top-Hat operator to solve the problem of uneven light and shade.After that,the improved phase spectrum of quaternion fourier transform algorithm is used to calculate saliency maps,extract target regions quickly,and remove interference from complex background environment.In view of the interference of islands and reefs in the target candidate area,we extract the shape and HOG features,design a voting method to carry out feature discrimination,and finally obtain the detection results.Experimental results have shown that the detection effect of the proposed method is better than that of ITTI and SR in complex environments,and the detection recall rate reaches92.83% in different environments.