Recognition Technology Research On Sea Ship Target Of UAV Remote Sensing Images

Ship target recognition technology have a very important practical significance and widerange of applications whether in civilian or military fields, such as maritime transport, marineresource monitoring, military reconnaissance and so on. It becomes a basic, quick andconvenient method to obtain the sea ship target information by using remote sensing technology.In this paper, We based on UAV aerial high resolution remote sensing ship image as a researchobject, proceed with the study from the remote sensing image object extraction, characterizationand feature matching, explore better methods and technologies for high accuracy recognition ofship targets. The main innovations of this dissertation are summarized as follows:(1) This paper proposes a new Fuzzy Gath-Geva hybrid method based on local informationof histogram to segment color image. The method provide accurate initial cluster centers forGath-Geva fuzzy cluster that can better carry out pixel clustering by using image histograminformation. Experimental results show that the algorithm can better segment the color image.(2) Based on Remote sensing image texture differences of seawater in the background, thepaper proposes an object extraction method on the color image segmentation. The algorithmsegment all target including water background, the finding out the sea background image whichis used to subtract the original image basing on the sea texture characteristic. Experiment inMATLAB environment verifies of the method can be better used for the extraction foregroundobject of remote sensing image.(3) Summary of the moments invariants theory and its application in target recognition. Thepaper analyze and compare Hu moments and Zernike moments that maintain image translation,scaling and rotation invariance on the basis of experiment, which mainly provide basis of shiptarget recognition for the next moments.(4) The paper provides an algorithm of ship targets recognition based on an improvedQuaternion Zernike moments. The traditional Zernike moments algorithm which change colorimage into gray image cause color information loss, so the paper directly change the RGB colorimage into a plural form by introduce the four element theory. In addition, the traditional Zernikemoments is not beneficial to reduce the dynamic range of an image, deficiency of informationredundancy, and so on, so this paper uses zero order geometric moments to execute datastandardization. Through theoretical proof and experimental verification, the algorithm can keepbetter target image translation, scaling and rotation invariance and can be applied to the unknown ship target recognition.