Spectrum Information hyperspectral imaging target detection technique
|School||Huazhong University of Science and Technology|
|Course||Communication and Information System|
|Keywords||Hyperspectral images Target Detection Logarithmic polar coordinate transformation Template matching Uncertainty Analysis|
In the field of pattern recognition, image-based object detection and recognition are always the most practical value and importance of the research direction, especially in the aerospace sector, it has become a key satellite navigation and positioning technologies. But the technology still has two defects, resulting in difficult to meet the actual requirements: (a) in target detection based on template matching technique, requires that the target with a pre-given reference templates are basically the same, but in reality, target point of view, even the gray scale distribution may be changed greatly; (2) based on spectral matching target detection technique, since the spectral feature inherent uncertainty can cause results to a large bias. This paper mainly studied the above two issues. Template correlation matching algorithm for real-time image occurs in large angle rotation and large scale changes in proportion, the positioning results may occur deviation problem, this paper proposes a logarithmic polar coordinate transformation based template matching improved method. The basic idea is: first through logarithmic polar coordinate transformation to obtain reliable target area, and as a new template diagram; Secondly, the use of a logarithmic polar coordinate transformation can change the size and rotation of the displacement characteristics, estimated logarithmic polar diagram template relative to the real-time target image offset, thus correcting the template graph. Finally, after the correction of the real-time graph traversal diagram template matching the target position determined. The proposed method for the existence of rotation, scaling, translation change objectives can automatically detect and identify the location. With Atlas hyperspectral image of unity features as remote sensing provides a new means. But in the actual case, since the atmosphere, sensors, location and other factors, coupled with the complexity of the feature itself, the impact of the spectral curve will make the feature mutate, creating uncertainty, resulting in \\In this regard, this paper based on the uncertainty of the spectral feature objective considerations, we propose a spectral-based target detection algorithm uncertainty. First of all these factors of uncertainty generated by spectral analysis of surface features of uncertainty and informed quantitative indicators, namely uncertainty; thus the uncertainty introduced into the spectrum detection algorithm in order to upgrade the traditional spectral recognition algorithm performance. Experimental results demonstrate the effectiveness of the proposed method.