计算机应用研究
計算機應用研究
계산궤응용연구
APPLICATION RESEARCH OF COMPUTERS
2013年
10期
3015-3021
,共7页
目标分类%主成分分析%Fisher线性判决%目标识别%飞行姿态识别
目標分類%主成分分析%Fisher線性判決%目標識彆%飛行姿態識彆
목표분류%주성분분석%Fisher선성판결%목표식별%비행자태식별
target classification%PCA%Fisher linear discrimination%target recognition%flight attituderecognition
利用3D-MAX建立了三种飞机模型, 设计了三维飞机目标图像库。在该图像库中按不同的飞机模型分为三个大类, 每个大类中按不同的飞行姿态分为24个组。论述了三类飞机图片的纹理特征的提取方法, 并通过基于贴近度的多传感器一致性融合方法计算每类飞机图像的对比度融合值、熵融合值、逆差距特征融合值, 通过最小距离法对测试对象的归类进行预判断。介绍了K-L变换、降维特征矩阵空间与投影系数向量, 计算了图库中每张图像在降维特征矩阵空间投影的系数向量, 以及测试图像在降维特征矩阵空间投影的系数向量。然后介绍了Fisher最佳鉴别向量与判别规则, 以图像在特征子空间投影的系数向量作为对象, 通过对每组图像正样本与负样本的训练得出最佳鉴别向量, 通过Fisher判别规则判断测试目标的类别与飞行姿态。通过实验证明了识别系统的高识别率, 以及可以对飞机复杂飞行姿态进行判断。
利用3D-MAX建立瞭三種飛機模型, 設計瞭三維飛機目標圖像庫。在該圖像庫中按不同的飛機模型分為三箇大類, 每箇大類中按不同的飛行姿態分為24箇組。論述瞭三類飛機圖片的紋理特徵的提取方法, 併通過基于貼近度的多傳感器一緻性融閤方法計算每類飛機圖像的對比度融閤值、熵融閤值、逆差距特徵融閤值, 通過最小距離法對測試對象的歸類進行預判斷。介紹瞭K-L變換、降維特徵矩陣空間與投影繫數嚮量, 計算瞭圖庫中每張圖像在降維特徵矩陣空間投影的繫數嚮量, 以及測試圖像在降維特徵矩陣空間投影的繫數嚮量。然後介紹瞭Fisher最佳鑒彆嚮量與判彆規則, 以圖像在特徵子空間投影的繫數嚮量作為對象, 通過對每組圖像正樣本與負樣本的訓練得齣最佳鑒彆嚮量, 通過Fisher判彆規則判斷測試目標的類彆與飛行姿態。通過實驗證明瞭識彆繫統的高識彆率, 以及可以對飛機複雜飛行姿態進行判斷。
이용3D-MAX건립료삼충비궤모형, 설계료삼유비궤목표도상고。재해도상고중안불동적비궤모형분위삼개대류, 매개대류중안불동적비행자태분위24개조。논술료삼류비궤도편적문리특정적제취방법, 병통과기우첩근도적다전감기일치성융합방법계산매류비궤도상적대비도융합치、적융합치、역차거특정융합치, 통과최소거리법대측시대상적귀류진행예판단。개소료K-L변환、강유특정구진공간여투영계수향량, 계산료도고중매장도상재강유특정구진공간투영적계수향량, 이급측시도상재강유특정구진공간투영적계수향량。연후개소료Fisher최가감별향량여판별규칙, 이도상재특정자공간투영적계수향량작위대상, 통과대매조도상정양본여부양본적훈련득출최가감별향량, 통과Fisher판별규칙판단측시목표적유별여비행자태。통과실험증명료식별계통적고식별솔, 이급가이대비궤복잡비행자태진행판단。
This paper established aircraft models by using 3D-MAX and designed a three-dimensional the aircraft target image library. In this image library, divided different aircraft models into three categories, each category were divided into 24 groups according to flight poses. It discussed the three types of aircraft image texture feature extraction methods. Then it calculated contrast fusion value, entropy of fusion, inverse gap value feature fusion of each type of aircraft image by multi-sensor consistency fusion method based on closeness, and pre-judge the test object classification by the minimum distance method. It also introduced K-L transform, feature subspace and projection coefficient vector, calculated the coefficient vector of each image's projection on gallery feature subspace, and tested image's projection on feature subspace. Then it introduced fisher optimal discriminate vector and discriminate rules. It considered coefficient vector of image projection on feature subspace as an object; derived optimal discriminate vectors through training each set of image's positive samples and negative samples. It judged test target's category and flight attitude through fisher discriminate rules. It proved a high recognition rate of the recognition system by experiments, as well as the judgment of the aircraft complex flight attitude.
