Font Size: a A A

Research On Small Target Detection For Hyperspectral Image Without Prior Knowledge

Posted on:2016-10-21Degree:MasterType:Thesis
Country:ChinaCandidate:K ChengFull Text:PDF
GTID:2308330464952782Subject:Detection Technology and Automation
Abstract/Summary:PDF Full Text Request
Because of the hyperspectral image data is much more than multi-spectral image’s, so it has stronger power on ground objects spectrum resolving, and people put more and more attention to the detection of the target based on hyperspectral information as supporting technology. However, the interested target is small target shown as a kind of abnormal data in ground objects. For this situation, we can adopt anomaly detection method to solve the problem. At present, this kind of method is one of the highlights in agro-scientific research in the hyperspectral remote sensing image processing. The paper is focus on some key technologies in the field of small target detection for hyperspectral image without prior knowledge and the main contribution can be summarized as follows:1. The relevant theoretical knowledge and characteristics of hyperspectral image are suggested. Among them, the article focuses on the characteristic of spectral correlation and spatial correlation. Then the theory of anomaly detection is analyzed. Besides, the algorithm of RX, local RX,PCA-RX are suggested and the experiments are conducted to compare their detection performance in the real hyperspectral image.2. The linear mixed model about hyperspectral image is analyzed. Among the technology of spectral unmixing, we focus on the algorithm of VCA and IEA. The experiments are also performed to compare their performance in the synthetic figure.3. The theory of LPD is analyzed and for making up for the shortcomings of the algorithm, a novel anomaly detection method based on improved LPD was presented. Firstly, the Iterative Error Analysis(IEA) algorithm is used to extract the endmembers and the endmembers which similar to the spectrum of background objects are chose to make up the background matrix. Then, the matrix is applied to build the orthogonal projection operator. Finally, the operator was used in the LPD to achieve the target detection. In order to check the results, we use the algorithm of Otus to split the gray image and remove the interfering targets by labeling the connected component. Experimental results show that our proposed method can restrain the background effectively and improve the detection performance obviously.
Keywords/Search Tags:Hyperspectral Image, Anomaly Detection, Low Probability Detection, Iterative Error Analysis, orthogonal projection operator
PDF Full Text Request
Related items