基于深度学习的地空导弹发射区拟合算法

目前地空导弹发射区的拟合算法主要是多项式拟合法和BP神经网络拟合法。多项式拟合法存在函数形式难以确定、函数范围不易分段等问题，且拟合精度较低；传统神经网络方法要想达到较高精度，需要大量的隐层节点，且在隐层节点数增加到一定程度后，训练变得十分困难且精度很难继续提高。同时，传统神经网络需要大量的标签数据，进一步增大了实际应用的难度。为此，基于深度学习理论，设计了一种基于堆栈稀疏自编码器（SSAE）的深度拟合网络（DFN），并给出了相应的训练策略。仿真实验表明其相比传统算法具有更小的拟合误差优势。所设计的深度稀疏自编码器网络可以克服多项式拟合和传统神经网络的不足，不仅可以在大量无标签数据和少量标签数据条件下进行学习训练，而且可以进一步提升地空导弹发射区的拟合精度。

Fitting algorithm of ground-to-air missile launching area based on deep learning

At present, the fitting algorithm for ground-to-air missile launching areas mainly include the polynomial fitting algorithm and the Back Propagation (BP) neural network fitting method. The former is problematic in that the function form is difficult to determine, the function range is not easy to segment, and the fitting precision is low, whereas the latter requires a large number of hidden layer nodes to achieve higher precision. When the number of hidden layer nodes increases to a certain extent, its training becomes very difficult and its precision is difficult to keep improving. At the same time, traditional neural networks require a large amount of labeled data, which further increases the difficulty of practical applications. So, this paper designs a Stacked Sparse Auto-Encoder (SSAE) Deep Fitting Network (DFN) based on the deep learning theory, and provides the corresponding training strategy. The simulation results show that the proposed design has the advantage of getting lower fitting error compare to traditional methods. The deep sparse auto-encoder network designed in this paper can overcome the shortcomings of polynomial fitting and traditional neural network. This design can learn training with a large amount of unlabeled data and a small amount of tag data, further enhancing the ground-to-air missile launching area fitting accuracy.