Detection of Navigation Marks in High resolution Remote Sensing Imagery

Abstract:

A novel method for extracting navigation mark using high resolution remote sensing imagery is proposed in this paper. The one class support vector machine(OCSVM) is used to segment the land and the water to derive the shoreline. Then the small targets within the water regions are found out and regarded as the candidate ones. The statistics of pixel intensity and the geometric feature of the candidate targets are used to remove a portion of false targets. Then the rest of the candidate targets are categorized into several groups according to the relationship coefficient between them and others. The group having most targets is the one that consists of navigation marks. At last, an online learning algorithm is proposed to decrease the miss rate. The spatial distribution of the extracted navigation marks are used to estimate the positions where the missing targets are likely to exist. The intensity distribution of the extracted navigation marks then are used as the prior knowledge to detect the missing target in the estimated positions. The experiments using QuickBird imagery show that the proposed method is effective.