系统工程理论与实践
繫統工程理論與實踐
계통공정이론여실천
Systems Engineering—Theory & Practice
2006年
4期
108~113
,共null页
特征提取 车辆识别 能谱密度 傅里叶变换
特徵提取 車輛識彆 能譜密度 傅裏葉變換
특정제취 차량식별 능보밀도 부리협변환
feature extraction; vehicle recognition; power spectral density; Fourier transform
声波和地震波是军事车辆类型识别的重要信息源,针对军事车辆运动时产生的声波和地震波,采用短时傅里叶变换提取其波形数据的频谱特征向量,提出基于能量频谱密度进行二次特征选择,构造声波和地震波频谱特征向量子空间,从而降低了特征向量的维数.应用支持向量机(SVM)和最近邻分类法(KNN)分别对声波和地震波数据来进行军事车辆分类,结果表明:基于能量频谱密度的二次特征选择方法能有效地构造出声波和地震波的特征子空间,由此得到的分类准确率高于传统的特征选择方法.通过比较SVM和KNN的分类结果可以得出SVM的分类效果优于KNN.
聲波和地震波是軍事車輛類型識彆的重要信息源,針對軍事車輛運動時產生的聲波和地震波,採用短時傅裏葉變換提取其波形數據的頻譜特徵嚮量,提齣基于能量頻譜密度進行二次特徵選擇,構造聲波和地震波頻譜特徵嚮量子空間,從而降低瞭特徵嚮量的維數.應用支持嚮量機(SVM)和最近鄰分類法(KNN)分彆對聲波和地震波數據來進行軍事車輛分類,結果錶明:基于能量頻譜密度的二次特徵選擇方法能有效地構造齣聲波和地震波的特徵子空間,由此得到的分類準確率高于傳統的特徵選擇方法.通過比較SVM和KNN的分類結果可以得齣SVM的分類效果優于KNN.
성파화지진파시군사차량류형식별적중요신식원,침대군사차량운동시산생적성파화지진파,채용단시부리협변환제취기파형수거적빈보특정향량,제출기우능량빈보밀도진행이차특정선택,구조성파화지진파빈보특정향양자공간,종이강저료특정향량적유수.응용지지향량궤(SVM)화최근린분류법(KNN)분별대성파화지진파수거래진행군사차량분류,결과표명:기우능량빈보밀도적이차특정선택방법능유효지구조출성파화지진파적특정자공간,유차득도적분류준학솔고우전통적특정선택방법.통과비교SVM화KNN적분류결과가이득출SVM적분류효과우우KNN.
Acoustic and seismic wave data play an important role in the recognition of military vehicles. We utilized the Short Time Fourier Transform (STFT) approach to extract the spectral feature vectors from the acoustic and seismic wave data of military vehicles. The power spectral density (PSD)-based feature selection method was proposed to reconstruct the feature subspace of the acoustic and seismic spectral vectors for decreasing the dimension of the feature vectors. Two type military vehicles were respectively classified by using the acoustic and seismic wave data. The classification results by Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) revealed that the PSD-based method of feature selection could represent the seismic and acoustic signals more efficiently in feature subspace and the accuracy is better than the traditional feature selection method which is obtained via directly feature-range cutting off. It also could be concluded that the effect of SVM to recognize the military vehicles is superior to that of KNN.