地球信息科学学报
地毬信息科學學報
지구신식과학학보
GEO-INFORMATION SCIENCE
2014年
6期
949-957
,共9页
秦善善%王世新%周艺%王福涛%刘文亮
秦善善%王世新%週藝%王福濤%劉文亮
진선선%왕세신%주예%왕복도%류문량
资源三号卫星%非下采样Contourlet变换%Gram-Schmidt变换%支持向量机%低空间分辨率全色影像%融合
資源三號衛星%非下採樣Contourlet變換%Gram-Schmidt變換%支持嚮量機%低空間分辨率全色影像%融閤
자원삼호위성%비하채양Contourlet변환%Gram-Schmidt변환%지지향량궤%저공간분변솔전색영상%융합
ZY-3 satellite%nonsubsampled Contourlet transform%Gram-Schmidt transform%SVM%low resolu-tion panchromatic image%fusion evaluation
资源三号卫星遥感数据是一种新型遥感影像，目前尚未有一种专门适用于资源三号卫星融合的方法。因此，提出了一种非下采样Contourlet变换（NSCT）与Gram-Schmidt（GS）变换相结合的融合方法，将非下采样Contourlet变换增强的空间信息用于补充具有高保真度Gram-Schmidt变换融合算法在影像清晰度方面的不足。采用线性回归的方法模拟低分辨率全色影像，将高分辨率全色影像、低分辨率全色影像及两者差值的细节影像分别进行非下采样Contourlet变换，对所得的高低频系数采取不同的融合策略进行自适应融合处理，得到新的全色影像。由低分辨率全色影像取代GS正变换第一分量，非下采样Contourlet变换得到的全色影像取代GS反变换第一分量，进行Gram-Schmidt正交变换，得到融合影像。与大多存在光谱扭曲的传统融合方法相比较，本文方法在光谱保真度、空间清晰度及地物分类精度方面都有明显的优势，说明该融合方法是一种适合资源三号卫星数据多光谱与全色影像融合的方法。
資源三號衛星遙感數據是一種新型遙感影像，目前尚未有一種專門適用于資源三號衛星融閤的方法。因此，提齣瞭一種非下採樣Contourlet變換（NSCT）與Gram-Schmidt（GS）變換相結閤的融閤方法，將非下採樣Contourlet變換增彊的空間信息用于補充具有高保真度Gram-Schmidt變換融閤算法在影像清晰度方麵的不足。採用線性迴歸的方法模擬低分辨率全色影像，將高分辨率全色影像、低分辨率全色影像及兩者差值的細節影像分彆進行非下採樣Contourlet變換，對所得的高低頻繫數採取不同的融閤策略進行自適應融閤處理，得到新的全色影像。由低分辨率全色影像取代GS正變換第一分量，非下採樣Contourlet變換得到的全色影像取代GS反變換第一分量，進行Gram-Schmidt正交變換，得到融閤影像。與大多存在光譜扭麯的傳統融閤方法相比較，本文方法在光譜保真度、空間清晰度及地物分類精度方麵都有明顯的優勢，說明該融閤方法是一種適閤資源三號衛星數據多光譜與全色影像融閤的方法。
자원삼호위성요감수거시일충신형요감영상，목전상미유일충전문괄용우자원삼호위성융합적방법。인차，제출료일충비하채양Contourlet변환（NSCT）여Gram-Schmidt（GS）변환상결합적융합방법，장비하채양Contourlet변환증강적공간신식용우보충구유고보진도Gram-Schmidt변환융합산법재영상청석도방면적불족。채용선성회귀적방법모의저분변솔전색영상，장고분변솔전색영상、저분변솔전색영상급량자차치적세절영상분별진행비하채양Contourlet변환，대소득적고저빈계수채취불동적융합책략진행자괄응융합처리，득도신적전색영상。유저분변솔전색영상취대GS정변환제일분량，비하채양Contourlet변환득도적전색영상취대GS반변환제일분량，진행Gram-Schmidt정교변환，득도융합영상。여대다존재광보뉴곡적전통융합방법상비교，본문방법재광보보진도、공간청석도급지물분류정도방면도유명현적우세，설명해융합방법시일충괄합자원삼호위성수거다광보여전색영상융합적방법。
ZY-3 satellite data, which is a new type of remote sensing imagery, has not yet owned a specific imag-es fusion method. This paper proposed a new fusion algorithm based on the nonsubsampled Contourlet transform and Gram-Schmidt (GS) transform according to the characteristics of CBERS-03 data. Firstly, the linear regres-sion method was used to generate the low-resolution panchromatic image with respect to the low-frequency pix-els. Then, the nonsubsampled Contourlet transform was applied respectively on the high-resolution panchromatic image, the low-resolution panchromatic image, and the image generated from their differences, to generate high frequency and low frequency coefficients. With these coefficients, a new panchromatic image was produced by using adaptive fusion strategies. At last, the fused image was obtained through GS orthogonal transformation, by replacing the first component of GS positive transformation with the low-resolution panchromatic image, and re-placing the first component of GS inverse transformation with the newly generated panchromatic image. Com-pared with the traditional image fusion methods, which always have defects of spectral distortion, the proposed method had obvious advantages in terms of spectral fidelity, spatial resolution and classification accuracy. In con-clusion, the proposed method is appropriate for images fusion on the ZY-3 multispectral and high resolution data.
