RUAN Zhongbo,CHEN Ze,ZHOU Liangchen,et al.A Neural Network Open Set Recognition Method[J].Journal of Chengdu University of Information Technology,2025,40(03):313-317.[doi:10.16836/j.cnki.jcuit.2025.03.009]
基于最小有向包围盒的神经网络信号开集识别方法
- Title:
- A Neural Network Open Set Recognition Method
- 文章编号:
- 2096-1618(2025)03-0313-05
- Keywords:
- neural network; open set recognition; kernel density estimation; minimum directed bounding box(OBB)
- 分类号:
- TP183
- 文献标志码:
- A
- 摘要:
- 神经网络存在固有缺陷,即在神经网络训练完成后,将训练集中没有的未知分类数据输入已经训练完成的神经网络时,会强行对未知分类数据进行已知分类而不是将其划分为未知分类。与传统的基于样本均值的算法不同,结合多维数据聚类算法提出一种基于核密度估计和最小有向包围盒算法的神经网络开集识别算法。算法能更好地表示神经网络输出特征量空间的密度分布,并在此基础上压缩特征量空间表示范围,提高分类未训练样本的准确率。实验证明,新算法在样本覆盖率100%的原始空间中的分类准确率达到98%,优于传统算法。
- Abstract:
- This paper is based on the inherent defect of neural networks: after the training of neural networks was completed, when the unknown classification data that is not available in the training set is input into the trained neural network, the network will force the unknown classification data to be known instead of dividing it into unknown classification. Different from the traditional algorithm based on sample mean, this paper proposes an unknown label classification algorithm based on kernel density estimation and minimum directed bounding box algorithm(OBB). This algorithm can better represent the density distribution of the output feature quantity space of the neural network, and on this basis, compress the representation range of the feature quantity space, and improve the accuracy of classifying untrained samples. Experiments show that the classification accuracy of the new algorithm reaches 98% in the original space with 100% sample coverage, which is better than the traditional algorithm.
参考文献/References:
[1] Zhu X,Goldberg A B.Introduction to Semi-Supervised Learning[J].Synthesis Lectures on Artificial Intelligence and Machine Learning,2009,3(1):1-130.
[2] Chapelle O,Zien A.Semi-supervised classification by low density separation[C].Tenth International Workshop on Artificial Intelligence and Statistics,2005:57-64.
[3] 张昊.基于深度学习的通信信号识别关键技术研究[D].成都:电子科技大学,2020.
[4] Joseph O’Rourke.Finding minimal enclosing boxes[J].International journal of computer & information sciences,1985,14(3):183-199.
[5] Gill Barequet,Sariel Har-Peled.Efficiently approximating the minimum-volume bounding box of a point set in three dimensions[J].Journal of Algorithms,2001,38(1):91-109.
[6] Stefan Gottschalk.Collision queries using oriented bounding boxes[C].Diss.The University of North Carolina,2000.
[7] Stefan Gottschalk,Ming C.Lin,Dinesh Manocha.OBBTree:A hierarchical structure for rapid interference detection[C].SIGGRAPH ’96 Proceedings of the 23rd annual conference on Computer graphics and interactive techniques.ACM,1996.
[8] ShengLi,YunFu.Robust multi-label semi-supervised classification[C].Big Data(Big Data),2017 IEEE International Conference on,2017,11-14.
[9] 张劭.基于深度学习的未知调制类型的信号识别[D].西安:西安电子科技大学,2018.
[10] Scheirer,Walter.Toward Open Set Recognition[J].IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE,2013,35(7):1757-1772.
[11] 周龙梅.基于深度学习的通信信号识别技术研究[D].成都:电子科技大学,2018:3-4.
[12] L F Maimó,A H Celdrán,M G Pérez,et al.Dynamic management of a deep learning-basedanomaly detection system for 5g networks[J].Journal of Ambient Intelligence and HumanizedComputing,2019,10(8):3083-3097.
[13] 吴莹.吴莹雷达辐射源指纹特征提取和识别技术研究[D].西安:西安电子科技大学,2018:6
[14] J Han,T Zhang,D Ren,et al.Mechanism analysis and feature extraction algorithm of commu-nication emitter fingerprint[J].AEU-International Journal of Electronics and Communications,2019,106:89-95.
[15] 贾永强.通信辐射源个体识别技术研究[D].成都:电子科技大学,2017:7-8.
备注/Memo
收稿日期:2023-11-28