计算机工程
計算機工程
계산궤공정
COMPUTER ENGINEERING
2013年
11期
163-168
,共6页
红外目标检测%核空间%特征提取%二次陒关滤波器%混合概率模型%子空间二次综合判别函数
紅外目標檢測%覈空間%特徵提取%二次陒關濾波器%混閤概率模型%子空間二次綜閤判彆函數
홍외목표검측%핵공간%특정제취%이차희관려파기%혼합개솔모형%자공간이차종합판별함수
infrared target detection%kernel space%feature extraction%Quadratic Correlation Filter(QCF)%mixture probabilistic model%Subspace Quadratic Synthetic Discriminant Function(SSQSDF)
针对二次陒关滤波器(QCF)与核空间特征陒结合在红外目标检测中的应用,提出 KSSQSDF 核直接映射法与 MPKPCA-SSQSDF核特征提取融合法。前者对低维空间下的QCF直接进行高维映射,使其转化为核空间下的非陑性陒关滤波器;后者采用核空间进行特征提取,对提取后的特征向量使用低维空间的陒关滤波器,用于红外目标检测。通过实验分析2种算法间的陒互联系,在目标检测结果及计算复杂性等方面的差异,结果表明,2种算法的检测精度大致陒同,均明显优于低维空间的QCF检测,但MPKPCA-SSQSDF核特征提取融合法不受QCF种类限制,检测时间短,具有广泛性,在某种程度上可以代替KSSQSDF核直接映射法。
針對二次陒關濾波器(QCF)與覈空間特徵陒結閤在紅外目標檢測中的應用,提齣 KSSQSDF 覈直接映射法與 MPKPCA-SSQSDF覈特徵提取融閤法。前者對低維空間下的QCF直接進行高維映射,使其轉化為覈空間下的非陑性陒關濾波器;後者採用覈空間進行特徵提取,對提取後的特徵嚮量使用低維空間的陒關濾波器,用于紅外目標檢測。通過實驗分析2種算法間的陒互聯繫,在目標檢測結果及計算複雜性等方麵的差異,結果錶明,2種算法的檢測精度大緻陒同,均明顯優于低維空間的QCF檢測,但MPKPCA-SSQSDF覈特徵提取融閤法不受QCF種類限製,檢測時間短,具有廣汎性,在某種程度上可以代替KSSQSDF覈直接映射法。
침대이차희관려파기(QCF)여핵공간특정희결합재홍외목표검측중적응용,제출 KSSQSDF 핵직접영사법여 MPKPCA-SSQSDF핵특정제취융합법。전자대저유공간하적QCF직접진행고유영사,사기전화위핵공간하적비이성희관려파기;후자채용핵공간진행특정제취,대제취후적특정향량사용저유공간적희관려파기,용우홍외목표검측。통과실험분석2충산법간적희호련계,재목표검측결과급계산복잡성등방면적차이,결과표명,2충산법적검측정도대치희동,균명현우우저유공간적QCF검측,단MPKPCA-SSQSDF핵특정제취융합법불수QCF충류한제,검측시간단,구유엄범성,재모충정도상가이대체KSSQSDF핵직접영사법。
Aiming at the Quadratic Correlation Filter(QCF) associated with kernel space is applied to infrared target detection, this paper proposes KSSQSDF kernel direct mapping algorithm and MPKPCA-SSQSDF kernel feature extraction fusion algorithm. KSSQSDF directly extends QCF from low dimensional space to high dimensional space, thus QCF is transformed to nonlinear correlation filter in kernel space. MPKPCA-SSQSDF first extracts target feature under kernel space, and then the extracted feature vector is used to QCF of low dimensional space for infrared target detection. Through experiment, the difference of detection result and computational complexity are analytically given when KSSQSDF and MPKPCA-SSQSDF are used respectively. The result shows kernel direct mapping algorithm and kernel feature extraction fusion algorithm have the similar detection accuracy, which evidently exceed QCF of low dimensional space. But MPKPCA-SSQSDF kernel feature extraction fusion algorithm does not confine the type of QCF, and has shorter detection time. So it has more extensive application range, and to some extent it can substitute for KSSQSDF kernel direct mapping algorithm.