Study On Aircraft Identification From High-resolution Remote Sensing Images Based On Object-oriented Analysis

With the advances in remote sensing technology, remote sensing image resolution is getting higher. Research on high-resolution image interpretation determines whether the image data can be efficiently and quickly serve the community. One of the important research areas as remote sensing image interpretation, research in remote sensing image target recognition technology is also facing new challenges and opportunities. In the field of target identification, the aircraft is normally used as an object of study. To accurately identify and locate the aircraft has important significance in both military combat and commercial investigation. High-resolution images rich in spatial, spectral, contextual information, which brings more details, but at the same time also brings a more complex background interference. In the face of massive details, there has been a decline in the efficiency and accuracy of the original, more mature target recognition algorithm suitable for the low-resolution image. When faced with this phenomenon that details in high resolution image is enlarged, relying solely on statistical analysis of pixel intensity values can not meet the requirements. Set of pixels gradually become the basic unit of study, and then the analysis concept based on object-oriented come into being, and it opens a new door for the analysis of high-resolution images. This paper studies the concept of object-oriented applications to achieve recognition of the aircraft in high resolution remote sensing image.In this paper, the concept of object-oriented analysis as a guide, summarizes the process for identifying high-resolution images of the aircraft. The first is the image preprocessing, in other words to get the object from the image; followed by objectoriented image recognition, that is, the extraction characteristics of the object, and on the basis of features to complete the identification task. The first step is to get a collection of objects, a process that includes an image segmentation, morphological processing, contour tracing, region topology connection, which will be repeated between the contour tracing and region topology connections to ensure the integrity of the primary image of the object. Proposed region topology connection method based on centroid direction of the connection, this method can repair cracks which exists between the fuselage and wing after segmentation, and maintain the integrity of the aircraft after the segmentation. The second step is to extract image features and image recognition, the process according to the basic geometrical characteristics of the object image on the basis of suspected target screening and advanced feature template matching. The whole process step by step removal of pseudo target is stepwise refinement process. The basic characteristics that area, aspect ratio, the degree of dispersion, the convex defect are mainly used. The advanced features of this paper include the Hu moments and the modified interest points feature. Experiments show that this set of methods, processes, features used, for different orientations and sizes of aircraft can achieve higher detection rate, false alarm rate can be suppressed at the same time, and in higher resolution have better performance.