基于泄露电缆的周界入侵数据增广方法研究

Research on Perimeter Intrusion Data Enhancement Method Based on Leaked Cable

机器学习的方法在泄漏电缆周界入侵检测领域有了比较好的发展。采集大量的入侵数据,通过机器学习训练模型,可以实现周界入侵检测更高的定位精度和更低的误报率。但是大规模入侵信号数据的缺乏限制了识别的准确率。而人工有效采集标注入侵信号数据极其费时且代价高昂。针对入侵信号数据缺乏的问题,本文引入了图像领域的数据增广方法。实验表明,几何变换增广结合深度卷积生成对抗的增广方法对周界入侵识别准确率的提升达到14.11%,极大程度上缓解了小样本下周界入侵定位精度低的问题,证明了本文增广方法的有效性。

Machine learning method has a good development in the field of leakage cable perimeter intrusion detection. By collecting a large number of intrusion data and training the model through machine learning, the perimeter intrusion detection can achieve higher positioning accuracy and lower false alarm rate. However, the lack of large-scale intrusion signal data limits the accuracy of identification. The manual and effective collection of labeled intrusion signal data is extremely time-consuming and expensive. Aiming at the problem of lack of intrusion signal data, this paper introduces the data augmentation method in the field of image. Experiments show that the augmented method of geometric transformation augmented combined with depth convolution generation countermeasure improves the accuracy of perimeter intrusion recognition by 14.11%, which greatly alleviates the problem of low perimeter intrusion location accuracy under small samples, and proves the effectiveness of the augmented method in this paper.