Research On The Method Of Digital Image Recognition Of Space Target

With the further development of space rush, the Earth Outer Space is graduallybecoming a new territory for military struggle. Space target recognition system plays afundamental and critical role in this new military struggle. Due to radar is limited for itshigh power and short detection distance, this dissertation took a research on the methodof digital image recognition of space target.According to the characteristic of the non-cooperative space target, this dissertationproposed a method of modelling3D space target based on several images andestablished the image library, which mainly solved the problem of modelling3Dnon-cooperative space target. The proposed method extracted a lot of structuralinformation from the limited pieces of the target images and analyzed the effect of theperspective projection, which greatly reduced the modelling error.This dissertation took full advantage of the SIFT(Scale Invariant FeatureTransform), optimizing the image library to feature vector library. The SIFT algorithmhas a good properties of translation invariance, rotation invariance and scale invariance.The SIFT features had been clustered by AP(Affinity Propagation) clustering algorithm.The feature vector library had been removed a lot of insignificant information, buildingup a good basis for fast target recognition.According to the simplex characteristic of space target image background, thisdissertation proposed a method of image segmentation through reducing the Otsuthreshold to partition the starry sky and the targets. Extracted the SIFT features from thesegmented contour image. Numerical experiments were presented to illustrate the goodperformance of proposed method by dealing with a great number of space images.Finally, according to the flow chart of target recognition, this dissertationestablished an interface of space target recognition under MATLAB circumstance. Theinterface could conveniently recognize the loaded image and show the recognition result.The experiments proved that the proposed method was fast and with a high correct rateof space target recognition.