提出了一种应用于红外目标仿真的模型。利用训练后的条件深度卷积生成对抗网络,只需输入随机噪声和类别标签,便能够自动产生预期类别的红外目标仿真图像。在手写数字数据集(MNIST)和红外数据集上分别训练模型参数,再进行自动生成实验,均可以产生高真实度的样本图像;将判别网络提取的特征用于分类实验,并将所提方法合成的图像用于数据增强,以提升分类器性能。研究结果表明,所提方法能够有效模仿红外辐射特征。

A model applied to the simulation of infrared targets is proposed. By the trained conditional deep convolutional generative adversarial networks, only the random noise and category label are necessary for the automatic generation of the simulation images of infrared targets belonging to the expected category. The parameters are trained on the handwritten digital dataset (MNIST) and the infrared dataset, respectively, and subsequently the automatic generation experiment is carried out, which can produce the high trueness sample images. The features extracted by the discrimination network are used in the classification experiments, and the images synthesized by the proposed method are used for data augmentation to improve the classifier performance. The research results show that the proposed method can effectively imitate the infrared radiation characteristics.