基于多传感器特征信息融合的目标识别技术研究

发布时间：2018-10-04 21:49

【摘要】：在现代化的战场中,由于基于单传感器的目标识别技术的自身局限性已经远远不能满足现在各领域的需求了,而基于多传感器多特征目标识别技术它可以综合利用各传感器的特性来完成更多的任务,基于多特征融合的识别技术可以在许多不确定环境和复杂背景下,通过将多个传感器的特征信息进行有效融合,不仅能够提高目标识别率,而且能减少有限的数据带宽。目前国内对基于多传感器特征信息融合的目标识别技术研究相对而言较少,所以非常迫切的等待对其展开深入而有效的理论和技术研究。为此,我们从以下几块内容去进行探讨和研究:(1)主要介绍了几种特征提取算法,通过外部传感器获得的数据,利用特征提取算法对其数据进行有效的数据提取,其中主要对Hu矩、仿射不变矩、小波矩和Zernike矩这四种主要特征提取算法进行了详细的分析和仿真。(2)主要研究了协方差矩阵算法,根据协方差矩阵它的特点,研究了基于协方差矩阵图像特征融合方法。通过构造有效可靠的区域协方差矩阵来对图像的特征信息进行融合,从而使融合后的特征信息具有更好地旋转不变性、尺度不变性,鲁棒性,类可分性强等特点。(3)主要对BP神经网络进行了分析和研究,针对其存在收敛速度慢和易陷入局部极小等问题,首先对BP神经提出了改进,通过采用附加动量来调整其权值和添加自适应学习算法选择学习率,然后通过理论分析提出并研究了运用粒子群算法去优化改进的BP神经网络的参数来进一步提高神经网络的学习速度,最后再利用上述的融合特征信息进行目标识别。实验结果表明:相对于传统的神经网络,利用粒子群来优化改进的BP神经网络在一定程度上式能够明显提高目标识别率,同时将该算法应用到了实际的识别系统中,证明了该算法的可行性和有效性。
[Abstract]:In the modern battlefield, because of the limitation of the target recognition technology based on single sensor, it is far from being able to meet the needs of various fields. Based on multi-sensor multi-feature target recognition technology, it can comprehensively utilize the characteristics of each sensor to complete more tasks, and multi-feature fusion based recognition technology can be used in many uncertain environments and complex background. By effectively merging the feature information of multiple sensors, not only the target recognition rate can be improved, but also the limited data bandwidth can be reduced. At present, the research on target recognition based on multi-sensor feature information fusion in China is relatively few, so it is very urgent to wait for the in-depth and effective theoretical and technical research. For this reason, we discuss and study the following parts: (1) this paper mainly introduces several feature extraction algorithms. Through the data obtained by external sensors, we use the feature extraction algorithm to extract the data effectively. Four main feature extraction algorithms, such as Hu moment, affine invariant moment, wavelet moment and Zernike moment, are analyzed and simulated in detail. (2) Covariance matrix algorithm is studied. The method of image feature fusion based on covariance matrix is studied. By constructing an effective and reliable regional covariance matrix to fuse the feature information of the image, the fused feature information has better rotation invariance, scale invariance and robustness. (3) the BP neural network is mainly analyzed and studied. Aiming at the problems of slow convergence rate and easy to fall into local minima, the improvement of BP neural network is put forward. By using additional momentum to adjust its weight and add adaptive learning algorithm to select learning rate, Then the particle swarm optimization algorithm is used to optimize the parameters of the improved BP neural network to further improve the learning speed of the neural network through theoretical analysis. Finally, the fusion feature information is used for target recognition. The experimental results show that compared with the traditional neural network, the improved BP neural network can improve the target recognition rate to a certain extent, and the algorithm is applied to the actual recognition system. The feasibility and effectiveness of the algorithm are proved.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP212

【相似文献】