摘要
Hopfield神经网络数据关联算法是密集多回波环境下一种有效的关联方法。但是,Hopfield神经网络大都使用S igmoid激活函数,主要缺点是函数在0点附近对噪声十分敏感。为克服上述缺点,引进了一种新的激活函数。新函数通过参数调节网络中各神经元对输入值的敏感区域,可以在降低对噪声的敏感度的同时,保持对信号的有效响应。仿真结果表明,在Hopfield网络中采用的新的激活函数,可以使NJPDA算法的均方根误差更接近JPDA算法,进一步提高了基于神经网络的关联算法的性能。
Hopfield Network joint probabilistic data association （NJPDA） Algorithm is an effective association method in dense multi - return environment. A shortage of sigmoid activation function in NJPDA is its sensitivity to noise when input is near 0. To conquer it, a new activation function（NAF） is introduced. By using a parameter controlled input sensitive region, NAF can reduce the sensitivity to noise, and response to signal efficiently. Simulation results show that using NAF can make NJPDA＇s RMS close to JPDA＇s, and improve the performance of NJPDA.
出处
《计算机仿真》
CSCD
北大核心
2009年第8期338-340,共3页
Computer Simulation
关键词
数据关联
神经网络
激活函数
Data association
Neural networks
Activation function
