Target Detection And Classification Based On Hyperspectral Image

Hyperspectral detection as a new type of remote sensing technology with highspectral resolution, which can be able to obtain the target reflectance properties indifferent bands,and achieve accurate extraction of the target detection andclassification by using the accurate extraction of the target spectral curve. This paperfocuses on the research of hyperspectral image processing technology, To achieve thetarget detection and classification based on the accurate extraction of the targetspectral curve.The research work and achievements include the following contents:1. Hyperspectral image preprocessing technology research: Mainly includes thehyperspectral image data dimensionality reduction, band removal and the number ofendmembers estimated. First remove the band with poor quality. Due to the highcorrelation between the adjacent bands of the hyperspectral image. So reduce the datadimension through orthogonal projection transform is necessary. Reducing thecorrelation between different bands so as to reduce the computational complexity, andfurther estimates the number of endmembers.2. Hyperspectral image endmember extraction and target detection technologyresearch. Firstly extract the target endmember of hyperspectral image using convexgeometry,.Then achieve the target detection in complex scenes combined withminimum energy constraint target detection technology.3. Hyperspectral unmixing technology research.Mixed pixels exist widely in theactual hyperspectral images, In this paper, hyperspectral image unmixing is realizedby using non-negative matrix factorization technique, and through adding thenon-negativity constrained, the minimum volume constrained promote decompositionconverges to the optimal value. Lead to better target detection and classificationresults.