卷积神经网络在炮兵对抗训练系统中的应用

本文针对炮兵对抗训练系统中炸点图像目标捕捉的问题，提出了一种基于YOLACT的炸点区域快速识别及分割方法。对特征提取网络结构和参数进行修改，结合预测分支网络和掩膜生成网络输出炸点位置和区域范围，根据区域信息得到炸点中心坐标。实验结果表明，在构建的炸点数据集上本文方法能准确地识别和分割炸点目标，速度达到21.2fps，整体上优于对比算法，较好地解决了炮兵对抗训练系统中的一个基本环节。

Application of CNN in Artillery Countermeasure Training System

A fast burst point area identification and segmentation algorithm based on YOLACT is proposed to capture the blast point in artillery countermeasure training system. Firstly, the feature extraction network structure and parameters are modified for the target of the blast point area. The prediction branch network and the mask generation network are combined to output the location and boundary area of the blast point. Finally, the location of the blast point is calculated according to the boundary information. The experimental results show that the method in this paper can accurately identify and segment the target of the blast point on the constructed blast point data set, and the speed reaches 21.2 fps, which is better than the comparison algorithm as a whole, and can solve a basic problem in the artillery confrontation training system.