神经网络在精确打击目标识别中的应用研究

摘要

目标识别正确与否直接影响到防空火力的部署、分配及有效打击.利用BP神经网络识别精确打击目标并进行了仿真实验.建立了样本库,提取图像的不变矩作为神经网络的输入量,分别采用基本梯度下降算法、有动量和自适应学习速率梯度下降算法和Levenberg-Marguardt优化算法训练BP网络.仿真结果表明,不变矩理论很好地解决了3维物体的2维图像在旋转、平移、缩放时能否成功提取图像特征的问题,而采用LM优化算法的BP神经网络训练速度快,识别准确率高.%Targets recognition directly affects the arrangement,distribution and effective attack of antiaircraft fire.BP neural network is used to recognize targets for accurate strike and a simulation experiment is made.Sample storage is set up; moment invariant of the pictures is taken as the input of the neural network.Basic gradient descent algorithm,gradient descent with momentum and adaptive learning rate algorithm and Levenberg-Marguardt (LM) optimization algorithm are employed to train BP network.The simulation results show that the moment invariant theory can resolve the problem of two-dimensional image character collection after revolution,translation and scaling.The BP neural network using LM optimization algorithm achieves a good classification precision and speed.