用均场逼近网络计算关联概率
COMPUTING ASSOCIATION PROBABILITIES USING MEAN-FIELD APPROXIMATION NETWORKS
-
摘要: 数据关联问题是密集多回波环境下多目标跟踪中的一个关键问题。在固定温度参数T=1下,通过构造适当的能量函数,使Boltzmann机演化达平衡状态后,各神经元状态的平均值即为所要求的关联概率的近似值。在此基础上,提出了用均场逼近网络快速计算关联概率的新方法。仿真结果验证了本文方法的有效性。
-
关键词:
- 数据关联; 神经网络; 均场逼近
Abstract: The data assciation problem is one of the key problems of multitarget tracking in dense multiple return environments. By constructing a suitable energy function, the average values of a Boltzmann machine (T = 1) are approximately equal to the association probabilities. Then, a new method for computing association probabilities using mean-field approximation network is presented. The simulations show that this method is effective. -
周宏仁,敬忠良,王培德.机动目标跟踪.北京:国防工业出版社,1991,第十三章.[2]Sengupta D, Iltis R. Neural solution to the multitarget tracking data association problem. IEEE Trans. on AES, 1989, AES-25(1): 96-108.[3]敬忠良等.基于随机神经网络的数据关联组合优化研究.控制理论与应用,1994,13(1): 257-263.[4]Zhou B, Bose N K. A comprehensive analysis of Neural solution to the multitarget tracking data association problem0. IEEE Trans. on AES, 1993, AES-29(1): 260-263.[5]Iltis R, Ting P-Y. Computing association probabilities using parallel Boltzmann machines. IEEE Trans. on NN, 1993, NN-4(2): 221-233.[6]Hinton, G. et al. Boltzmann machines: Constraint satisfaction networks that learn. 1984, CMU-SC-84-119.[7]Ackley D, et al. A learning algorithm for Boltzmann machines[J].Cognitive Sci.1985, 9(1):147-169[8]Kong S Y. Digital Neural Network. New York: Prentice Hall, 1993, Ch.9.[9]焦李成.神经网络计算.西安:西安电子科技大学出版社,1993, 107-110. -
计量
- 文章访问数: 2156
- HTML全文浏览量: 134
- PDF下载量: 414
- 被引次数: 0
下载:
