A Study Of Hyperspectral Image Target Detection Technology Based On The Map Information And Spectrum Information

In the field of pattern recognition, image-based object detection and recognition are always the most practical value and importance of the research direction. Particularly in the aerospace field, it has become a key satellite navigation technology. However, there are still two major shortcomings of the technology causing it doesn't meet the requirements in practice. The first shortcoming is that the template matching based method requires that the given target and reference images should be consistent in their sizes; however, there are certain difference between the target and template images since their angles and scales even the gray may be changed to certain exten; the second problem is that for the target detection techniques based on spectral matching, it will cause results to a large deviation due to the uncertainty of the spectrums. In this paper, the traditional algorithms for the above two issues were studied.For the detection method based on the template matching technique, in order to overcome the disadvantage casued by the variations of angles and sizes, an improved template matching method based on the Log-polar transform was proposed. The entire schme of the proposed method was consist the following components: firstly, by Log-polar transform, the matching relibale region was obtained and used for the new template map; secondly, using the characteristics of Log-polar transform which can transform the size and rotation into the displacement in the Log-polar field, to estimate the bias between the matched image pair; and then the real image was corrected by the above bias. The final matching result was obtained by the corrected real image in the relabile region.Different from the traditional detecting technique, hyperspectral remote sensing not only provides 2D images of space, but also the spectral information. The fact that different objects have different spectrums was the foundation of the image classification based on spectrums. However, in practice, the object spectrum was not stationary because of the influence of geographical position, sensors, environment, and so on, which casued the phenomena of"same object with different spectrums"and"different objects same image". The uncertainty of spectrum deduced the performance of the image calssion method based on spectrum. Considering the impact of uncertainty on the spectrum uncertainty, an newly detection method was proposed. By analyzing the spectrum untainty, and quantitative indicators of spectrum uncertainty was described. Thus by introducing the uncertainty into the traditional spectral detection algorithms, an modified detection method was designed. Theorectical analysis shows that the execllent performance of the proposed methed. And several experimental results also demonstrate the effectiveness of the proposed method.