摘要
传统的特征值检测法需要计算信号检测统计量和检测阈值,在不同的应用场景下,需要运用不同的特征提取算法来提取信号特征,却难以找到合适的特征提取算法。基于此文中提出一种改进的基于特征向量的K-D树KNN-SVM联合分类器算法。该算法相比传统的特征值检测法,无需计算检测统计量和检测阈值,且在KNN部分将训练样本排列成K-D树结构,可以大大减少KNN部分的冗余计算,使得支持向量机在分类超平面模糊时,加快K近邻算法的搜索速度。仿真实验结果表明,在选定最佳参数的K-D树KNN-SVM联合分类器中,相比KNN或SVM频谱感知算法,其检测性能明显提高,且检测效率也比KNN-SVM高。
For the traditional eigenvalue detection method,signal detection statistics and detection threshold need to be calculated.In different application scenarios,different feature extraction algorithms are needed to extract signal features.However,it is difficult to find them.On this basis,an improved K⁃D(k dimensional)tree KNN⁃SVM(k nearest neighbor and support vector machine)joint classifier algorithm based on feature vectors is proposed in this paper.In comparison with the traditional eigenvalue detection method,it does not need to calculate the detection statistics quantity and detection threshold,and in the KNN part,to arrange the training samples into K⁃D tree structure can greatly reduce the redundant calculation,which can speed up the search speed of KNN algorithm in the classification of hyperplane fuzziness.The results of the simulation experiment show that the detection performance and efficiency of K⁃D tree KNN⁃SVM joint classifier with optimal parameters are all higher than KNN or SVM spectrum sensing algorithm.
作者
蒋礼君
张晓格
JIANG Lijun;ZHANG Xiaoge(Nantong Research Institute Company Limited for Advanced Communication Technologies,Nantong 226019,China;School of Information Science and Technology,Nantong University,Nantong 226019,China)
出处
《现代电子技术》
2021年第16期7-13,共7页
Modern Electronics Technique
基金
国家自然科学基金(61871241)
南通大学智能信息技术联合研究中心开放课题(KFKT2017A05)。
关键词
联合分类器
频谱感知
认知无线电
特征提取
机器学习
冗余计算
K-D树结构
joint classifier
spectrum sensing
cognitive radio
feature extraction
machine learning
redundant calculation
K⁃D tree structure
