CAJ | 학술논문

GF-1号卫星是中国高分卫星系列首颗卫星，自2013年04月26日发射以来，提供了大量的2 m/8 m/16 m空间分辨率的卫星数据，成为中国农业遥感监测的主要数据源之一。该文以GF-1卫星携带的16 m空间分辨率的宽视场（wide field view，WFV）传感器为主要数据源，采用2013年10月2日、10月17日、11月7日和12月5日4个时相的数据，以多尺度分割后的对象为基本分类单元，采用分层决策树分类的方法对冬小麦面积进行提取，并利用地面样方数据对分类结果进行了精度验证。结果表明，北京市顺义区冬小麦面积7095 hm2，分类总体精度达到96.7%，制图精度为90.0%，其他未分类类别精度为97.3%，Kappa系数为0.8。研究区内冬小麦的播种时间可以分为10月1－5日早播、10月6－10日中播、10月11－15日中晚播、10月16－20日晚播等4个时间段，不同播期对应着归一化植被指数（normalized difference vegetation index，NDVI）不同的变化规律，是分层的基础，结合波段反射率、波段反射率和、波段反射率比值等参数的变化规律，通过分层可以有效的剔除草坪、桃树等容易同冬小麦混淆的地物类型，GF-1/WFV 提供的多时相遥感数据能够可靠的反映冬小麦发育变化的规律，是冬小麦面积准确提取的基础，在农作物面积遥感监测业务运行中具有较大的开发应用潜力。
GF-1호위성시중국고분위성계렬수과위성，자2013년04월26일발사이래，제공료대량적2 m/8 m/16 m공간분변솔적위성수거，성위중국농업요감감측적주요수거원지일。해문이GF-1위성휴대적16 m공간분변솔적관시장（wide field view，WFV）전감기위주요수거원，채용2013년10월2일、10월17일、11월7일화12월5일4개시상적수거，이다척도분할후적대상위기본분류단원，채용분층결책수분류적방법대동소맥면적진행제취，병이용지면양방수거대분류결과진행료정도험증。결과표명，북경시순의구동소맥면적7095 hm2，분류총체정도체도96.7%，제도정도위90.0%，기타미분류유별정도위97.3%，Kappa계수위0.8。연구구내동소맥적파충시간가이분위10월1－5일조파、10월6－10일중파、10월11－15일중만파、10월16－20일만파등4개시간단，불동파기대응착귀일화식피지수（normalized difference vegetation index，NDVI）불동적변화규률，시분층적기출，결합파단반사솔、파단반사솔화、파단반사솔비치등삼수적변화규률，통과분층가이유효적척제초평、도수등용역동동소맥혼효적지물류형，GF-1/WFV 제공적다시상요감수거능구가고적반영동소맥발육변화적규률，시동소맥면적준학제취적기출，재농작물면적요감감측업무운행중구유교대적개발응용잠력。
GF-1 Satellite is the first one of the high resolution satellite series in China. Since its launch on April 26, 2013, GF-1 Satellite has provided a large amount of satellite data with high spatial resolutions of 2, 8 and 16 m, and it has become one of the major data sources for agricultural remote sensing monitoring in China. By taking WFV (wide field view) Sensor carried on GF-1 Satellite with the spatial resolution of 16 m as its major data source, using the data of 4 time phases, i.e. October 2, October 17, November 07 and December 05, 2013, and taking the objects after multi-resolution segmentation as its basic classification units, the paper extracts the winter wheat area by employing hierarchical decision tree classification method, and verifies the accuracy of the classification result by using the ground sample data. The result shows that, the total winter wheat area in Shunyi District, Beijing City is 7 095 hm2, with the overall classification accuracy of 96.7% and mapping accuracy of 90.0%. Accuracy of other unclassified types is 97.3%, with the Kappa coefficient of 0.8. The sowing period of winter wheat in the study area is classified into 4 sowing types: Early sowing (October 1st-5th), mid-term sowing (October 6th-10th), mid-late sowing (October 11th-15th) and late sowing (October 16th-20th). It is found that the NDVI (normalized difference vegetation index) values of winter wheat in above 4 sowing periods show a changing pattern of high-low-secondary high-high, which is closely associated with the development features of winter wheat. The higher the NDVI value on October 2nd, the later the sowing period of winter wheat will be, and the higher the NDVI value on December 5th, the earlier the sowing period will be. The change of NDVI value of late sowing winter wheat is the most significant. Under the support of ground training samples, the threshold range of NDVI is classified, and the 4 winter wheat’s sowing periods, i.e. early, mid-term, mid-late and late sowing are corresponding to different NDVI levels. With the NDVI values of different levels not overlapping, the paper calculates 32 parameters of 4 types, such as the reflectivity in Waveband 1-4, the sum total of the reflectivity of Waveband 1-4, the ratio between Waveband 4 and 3 and the ratio between Waveband 3 and 2. The threshold values of the 32 parameters are sequentially screened by employing decision tree classification method. Decision tree process includes the following steps: 1) To set up step length of 32 parameters; 2) To randomly select 10% of the step length combination; 3) To calculate the decision results of each combination; 4) To verify the accuracy of the results by relying on 10 training samples; 5) To select the combination with the highest accuracy as the threshold value of a decision tree node. Multi-temporal remote sensing data provided by GF-1/WFV can reliably reflect the changing law of winter wheat development. By data layering, the ground object types which are easy to be confused with winter wheat, such as grass lawn and peach tree, can be effectively eliminated, and the data can be taken as the foundation for accurate extraction of winter wheat area. Thus, GF-1/WFV has great development and application potential in remote sensing monitoring operations for crop area.