基于1DC-CGAN和小波能量特征的引信小样本地形目标识别

无载波超宽带引信由于具有定距精度高、抗截获能力强、穿透性好、有一定反隐身能力等特点，在多个弹药平台上得到应用。在对地面目标作用时，不同地形会影响引信炸高，从而影响毁伤效果。首次提出将无载波超宽带引信用于地形识别，为引信自适应确定最佳炸高提供先决条件。地形回波的采集周期长、成本高，获取回波的数量往往较少，这会影响识别精度。为扩充数据集，提出一种改进的条件生成对抗网络，将生成器和判别器的全连接层替换为一维卷积同时增加批标准化，在实现信号生成的同时减小模式崩溃问题发生的概率，提升了小样本条件下的序列生成效果。将扩充回波信号的小波能量特征作为输入特征，利用粒子群优化的反向传播(PSO-BP)神经网络实现地形智能分类。实验结果表明：相比在原始训练集上训练，扩充训练集上训练的PSO-BP神经网络在测试集上取得了4%以上的准确率提升。

Identification of Fuzzy Small-sample Terrain Targets Based on 1DC-CGAN and Wavelet Energy Features

The carrier-free UWB fuze is featured by high distance resolution, strong anti-interference capability, and rich information about target structure. Also, it is not easily affected by light and climate conditions. When striking ground targets, different terrain will affect the blast height of the fuze, which in turn affects the damage effect. A terrain identification system based on carrier-free UWB fuze is thus proposed, which requires rich experimental data for accurate target identification. The acquisition of terrain echoes is time-consuming and costly, and the number of acquired echoes is often limited, which may affect the recognition accuracy. To expand the data set, an improved conditional generation adversarial network is proposed, replacing the fully connected layers of the generator and discriminator with one-dimensional convolution, adding batch normalization to achieve signal generation while reducing pattern collapse, and enhancing the sequence generation effect under small sample conditions. In addition, the wavelet energy features of the expanded echo signals are used as input features, and the particle swarm optimized BP (PSO-BP) neural network is used to achieve intelligent terrain classification. Experimental results show that the PSO-BP neural network trained on the expanded training set has improved the accuracy by more than 4% compared with training on the original training set.