Hyperspectral Target Detection Algorithm With Spatial Information

Hyperspectral remote sensing is a head area in current remote sensing technology, it uses a lot of narrow electronica bands to require continuous diagnostic spectral characteristics from ground materials of interest, providing a reliable measurement to ground materials precision target detection. However, since spatial resolution of hyperspectral remote sensing images is always lower than that of multispectral images in the same imaging condition, the targets of interest in the hyperspectral images are in a few pixels or less than one pixel. This brought difficulties doing target detection with hyperspectral images.In this thesis, based on traditional hyperspectral target detection algorithms, we considered spatial information of targets to constraint spectral-dimensional detection, designed and realized an improved target detection algorithm based on spatial information. And one step further, we combined progressive band processing real-time target detection method, and explored the progressive band processing real-time spatial-spectral target detection algorithm. Through this research, we came up following conclusions:(1) The physical foundation and crucial problems of hyperspectral remote sensing target detection were researched, and the main stream hyperspectral target detection algorithms were analyzed in detail. And experiments were designed in both with and without simulate Gaussian noise ways to compare and judge the performance of different algorithms.(2) The spatial information based target detection algorithm come up in this paper can eliminate the influence of target pixels in global statistical estimation of background distribution model, and make the background statistical model more precisely, to achieve better background suppression and improve the precision of traditional spectral-dimensional target detection algorithm. And since the priori spatial information was considered, the anomaly detection algorithm had an ability to distinguish normal anomalies and targets of interest in some level, reducing the false alarm rate effectively, and made the anomaly detection algorithm used for target alarm of special kind of target more pertinence. This is helpful for hyperspectral anomaly detection spreading in real applications.(3) Progressive band processing real time spatial-spectral target detection algorithm keeps the processing mechanism of origin detection algorithm, and the detection result is the same as the original algorithm as long as the input band number is the same. And the real time spatial-spectral anomaly detection algorithm can integrated in the hardware to accomplish real time anomaly task, improving the time effectiveness of target detection and the rapid response of the detection result.