基于证据理论的双传感器决策融合目标检测

Target detection by decision-level fusion of dual-sensor based on evidence theory

针对单传感器图像目标检测概率相对较低的问题,提出用来自两个传感器的图像分别进行目标检测,并基于加权证据理论将检测结果进行决策级融合。在对同一地区SAR图像和高光谱图像进行决策级融合过程中,各证据的权值确定采取了以传感器信任度决定权值的方法,实现了各传感器之间图像信息最优化互补。实验结果表明,在虚警概率为10-3数量级的条件下,加权融合后的检测概率达到84.51%,比仅用单一高光谱图像和SAR图像进行目标检测时分别提高了11.27%和19.72%;在主观视觉效果上,采取决策级融合后检测效果也更好。

As for the problem that the target detection probability of single sensor image is relatively low, it is proposed that target detection is implemented separately by using images from two sensors, and decision-level fusion of detection results is accomplished based on weighted evidence theory. In the process of decision-level fusion from SAR image within the same district and hyper-spectral image, the weights of evidences are determined by the trust factor of sensors, fmally implemented optimized complementation of image information between sensors. As experiments show, guaranteeing the false alarm conditions in magnitude of 103, the probability of detection reaches 84.51% after weighted image fusion, higher than single hyper-spectral images and SAR images for target detection, increased by 11.27% and 19.72% each; in the subjective visual effects, the detection results after decision-level fusion are also better.