The Oil Depot Detection And Extraction From High Resolution Remote Sensing Images

At present,the development of satellite remote sensing technology has promoted the progress and development of other related fields.The resolution,recognition,effectiveness and other aspects of remote sensing images have been greatly improved by leaps and bounds.Remote sensing images have been applied to military,agricultural,meteorological,hydrological,environmental monitoring,disaster warning etc.Especially,they are particularly important for military applications with obvious advantages,such as targets reconnaissance of remote sensing satellites without any restrictions.Nowadays,as an important strategic targets,the geographical locations and strategic layouts of oil depots are very important.The detection and finding of the distribution of oil depots in time have comparatively significant strategic meaning.If the enemy oil depots were destroyed at first time,the initiative in the war will be grasped firmly by us.Therefore,the detection and extraction of oil depots in high resolution remote sensing images are very important for national security and national defense.This paper focuses on the detection and extraction of oil depots in high resolution remote sensing images.For finding oil depots in high resolution remote sensing images quickly and accurately,an extraction method of oil depots is proposed.It is a fusion of deep learning,saliency detection,image segmentation,Hough transform,and depth-first-search algorithm.Firstly,the region of interest(ROI)of oil depots in high resolution remote sensing images are detected rapidly using deep learning.After that,there are probably about 20% false alarms which should be excluded by the following steps.LC is a famous visual saliency detection method,and it is used to measure the saliencies of ROIs.Then,image binarization threshold segmentation is implemented to eliminate most interference factors.Finally,Hough transform circle detection is used to find out the center locations of circular targets,and the depth-first search algorithm is further used to detect and extract the oil depots sophisticatedly according to the distribution of oil depots.A dataset of high resolution remote sensing images containing oil depots is constructed,and these images are collected from GF-2 satellite of China and Google Earth.The experiments on this dataset have shown that the proposed method has improved both the precision and recall rate compared with the improved Hough transform detection algorithm.The precision reached 88.9%,and the recall rate reached 89.7%.The proposed method has greatly improved the detection accuracy and efficiency of oil depots in high resolution remote sensing images.