Research On Hyperspectral Imagery Anomaly Detection Based On Saliency

With the progress of imaging spectrometers and their data processing technologies,the related techniques for target detection using hyperspectral data acquired by imaging spectrometers have been greatly developed as well.Anomaly detection has become one of the research hotspots.It is a target detection method that does not require prior information and is non-supervised.Therefore,it is really practical.This paper aims at the shortcomings of the traditional anomaly detection algorithm,combining with the visual saliency method in computer vision,the traditional anomaly detection algorithm is improved.We use synthetic and real hyperspectral data for experiments.The main content of the paper includes:1.The basic concepts of anomaly detection in hyperspectral images are introduced.Moreover,the development status of anomaly detection and visual saliency is analyzed.2.The data format and data description model of hyperspectral image are briefly introduced.The concept of hyperspectral anomaly detection and the classic anomaly detection algorithms are described in detail.Furthermore,the evaluation criteria of anomaly detection are given.3.Aiming at the shortcomings of the traditional anomaly detection algorithm,a hyperspectral image anomaly detection algorithm based on context-aware saliency is proposed.By introducing the context-aware saliency model,the background modeling method of the image is improved.A weight map based on context-aware saliency is established,meanwhile the mean vector and covariance matrix describing the background in the conventional algorithm are redefined to achieve the purpose of optimizing the background estimation.The experiment results show that the algorithm greatly improves the detection efficiency compared with the traditional algorithm,but the calculation speed of the algorithms is really slow.4.Aiming at the problem of slow computation speed of the algorithm,a hyperspectral image anomaly detection algorithm based on spectral residuals saliency is proposed.The spectral residual saliency model is introduced and the saliency map is obtained by applying the Fourier transform method in the frequency domain.The experiment results show that the algorithms has a rapid calculation speed while improving the detection effect.In this thesis,by analyzing the shortcomings of the traditional anomaly detection algorithms,two hyperspectral image anomaly detection algorithms based on visual saliency are proposed.For the actual application of the detection effect and the calculation speed,the two algorithms have their own advantages.