基于多级神经网络的类型融合

研究了基于多级神经网络的类型融合方法。这种多级神经网络分为传感器子网和融合子网两部分。传感器子网是一种基于专家规则的模糊神经网络,根据专家规则确定网络结构,网络节点和传递函数都有明确的意义,避免了普通神经网络层数和隐层节点数难以确定的缺点。经过训练的传感器子网能够实现各目标类型的置信度分配,然后用融合子网对多个传感器子网输出结果进行融合,得到目标类型的最终判决。在融合子网中,加入了各传感器的可信度,使融合结果更可靠。仿真结果表明,此方法鲁棒性强,识别率高。

Type fusion based on multistage neural network

The target type fusion method based on multistage neural network is studied.This kind of multistage neural network is composed of sensor subnet and fusion subnet.The sensor subnet is a neural network based on expert rules,which construct its structure according to the expert rules.This neural network overcomes the drawbacks of the common neural network which has difficulty in ascertaining the number of net layers and hidden layer nodes.The sensor subnet which has been trained could obtain the likelihood of the type of every target and then the output of the sensor subnet could be fused by fusion subnet.The confidences of sensors are considered in fusion subnet and make the results of fusion more credible.Simulation results show that the method is effective.