Remote sensing change information detection and recognition has been a technical dififculty of remote sensing dynamic monitoring. This paper takes the east of Dongting Lake as the study area, GF-1 remote sensing image as the research object, on the basis of data pretreatment, divided study area into reed, sedge, Polygonum hydropiperand mud Artemisia, water, mud with 6 types. This study introduce of NDVI vegetation index and theifrst principal component band (PC1) to improve the traditional image difference algorithm and extract the change information of two images to compare with detection algorithm of support vector machine classification of multitemporal images. The results showed that: (1)Remote sensing images after atmospheric correction and image registration processing, optimum band combination of GF-1 remote sensing image change detection is RGB=432; (2) Using support vector machine classiifer to classify the remote sensing image, sample selection degree of isolation were between 1.9-2.0 The overall accuracy of classiifcation results is 85.34% and Kappa coefifcient is 0.8 that meet the post classiifcation comparison algorithm to extract change information requirements; Introduction of NDVI and theifrst principal component to distinguish the change information, using the histogram accumulation interval to confirm change threshold,the result shows that the optimization result is the best when change threshold of increase information is set to 0.3, change threshold of decrease information is set to 0.2 , the Smooth Kernel Size is set to3 and aggregation Min Size is set to 30. Extraction the change information of wetland from remote sensing images of GF-1, compared with the improved image difference algorithm after classiifcation, the image difference method can quickly, directly extract change information and the result is not affected by classiifcation accuracy and sample consistency . The detection precision is 89.6%.Kappa coefifcient is 0.9. Image difference algorithm is superior to the traditional classiifcation algorithm and it is an efifcient and feasible method.%湿地遥感变化信息检测并识别一直是遥感动态监测的一个技术难点。以东洞庭湖为研究区，2期GF-1遥感影像为研究对象，在数据预处理的基础上，将研究区分为芦苇、苔草、辣蓼与泥蒿、水体、泥滩地等6种类型。研究引进了NDVI植被指数波段与第一主分量波段（PC1）对传统的图像差值算法进行改进，提取出两期影像的变化信息，并与支持向量机的多时相影像分类后检测算法相比较。结果表明：（1）研究区遥感影像经过大气校正和图像配准等预处理之后，GF-1遥感影像变化检测的最佳波段组合为RGB=432；（2）利用支持向量机分类器对两期遥感影像进行分类时，样本选择的可分离度均在1.9～2.0之间，分类结果的总体精度为85.34%，Kappa系数为0.8，满足分类后比较算法提取变化信息的要求；（3）引进NDVI与第一主分量区分变化信息，并采用直方图积累区间确定变化阈值，信息增加的变化阈值设置为0.3，信息减少的变化阈值设置为0.2，Smooth Kernel Size设置为3，Aggregation Min Size设置为30，优化结果最佳。利用2期GF-1遥感影像提取湿地变化信息，分类后比较算法与改进后图像差值算法，图像差值法快速、直接提取变化信息，检测精度为89.6%，Kappa系数为0.9，且不受分类精度与分类样本一致性的限制，明显优于传统分类比较算法，是一种高效可行的方法。
展开▼
