基于特征增强的辐射源开集识别方法

Feature-enhanced open-set recognition of specific emitter

摘要:

现有的辐射源个体识别研究面临着真实开集环境下信号特征判别能力弱、特征边界模糊等挑战。为此,提出了一种基于特征增强的辐射源开集识别方法。针对辐射源信号特征辨别能力弱的问题,设计了基于三重Sigmoid函数的网络模块,提取更丰富、更具代表性的特征,提升辐射源信号的特征表示能力。针对特征边界模糊的问题,引入了能量模型来计算辐射源数据的能量损失,并联合交叉熵损失和三元损失函数,对特征边界阈值进行优化。实验结果表明,该方法在辐射源数据集上的AUC评估指标优于现有的前沿开集识别方法,并且在特征空间中有效区分了不同类别的特征边界,显著提升了辐射源开集识别的准确率。

Abstract:

The existing research on specific emitter identification faces challenges such as weak signal feature discrimination and blurred feature boundaries in real open-set environments. To address these issues, this paper proposed a feature-enhanced open-set recognition method for specific emitter. To improve the weak discriminative ability of specific emitter signal features, a network module based on the Triple Sigmoid function was designed to extract richer and more representative features, significantly enhancing the feature representation capability of specific emitter signals. To tackle the issue of blurred feature boundaries, an energy model was introduced to calculate the energy loss of specific emitter data, combined with cross-entropy loss and triplet loss functions to optimize the feature boundary thresholds. Experimental results show that the proposed method outperforms current state-of-the-art open-set recognition methods in terms of AUC on specific emitter datasets and effectively distinguishes feature boundaries of different classes in the feature space, significantly improving the accuracy of open-set specific emitter recognition.