CAJ | 학술논문

通过对长时间序列遥感影像的波谱变化特征分析，可以有效地进行农作物种类识别与信息提取，提高农作物种植面积的遥感监测精度。中空间分辨率多光谱遥感影像适合于中国大范围大宗农作物面积监测，也是能够提供稳定时间序列遥感数据源之一。该研究以河北省衡水市为研究区域，采用2011年10月3日－2012年10月24日期间，16景30 m空间分辨率的HJ-1A/B卫星CCD（电荷耦合元件，charge-coupled device）影像月度NDVI（归一化植被指数，normalized difference vegetation index）时间序列数据，针对冬小麦、夏玉米、春玉米、棉花、花生和大豆等主要作物类型，在全生育期波谱特征曲线分析基础上，提取主要作物类型的曲线特征，采用基于 NDVI 阈值的决策分类技术，进行了农作物种植面积遥感识别，以15个规则的2 km×2 km的地面实测GPS（全球定位系统，global positioning system）样方进行了精度验证。考虑到大豆和花生2种作物的NDVI时间序列特征相似性较高，将这2种作物合并为一类进行分类，并命名为小宗作物。结果表明，冬小麦、夏玉米、春玉米、棉花和小宗作物等5类目标可以有效识别，分类总体精度达到90.9%，制图精度分别为94.7%、94.7%、82.4%、86.9%和81.2%，其他未分类类别精度为85.9%。利用中高分辨率遥感时间序列卫星影像，在大宗农作物时间序列的变化规律分析基础上，可以准确地提取大宗农作物种植面积，在农作物面积资源调查中具有较大的应用潜力。
통과대장시간서렬요감영상적파보변화특정분석，가이유효지진행농작물충류식별여신식제취，제고농작물충식면적적요감감측정도。중공간분변솔다광보요감영상괄합우중국대범위대종농작물면적감측，야시능구제공은정시간서렬요감수거원지일。해연구이하북성형수시위연구구역，채용2011년10월3일－2012년10월24일기간，16경30 m공간분변솔적HJ-1A/B위성CCD（전하우합원건，charge-coupled device）영상월도NDVI（귀일화식피지수，normalized difference vegetation index）시간서렬수거，침대동소맥、하옥미、춘옥미、면화、화생화대두등주요작물류형，재전생육기파보특정곡선분석기출상，제취주요작물류형적곡선특정，채용기우 NDVI 역치적결책분류기술，진행료농작물충식면적요감식별，이15개규칙적2 km×2 km적지면실측GPS（전구정위계통，global positioning system）양방진행료정도험증。고필도대두화화생2충작물적NDVI시간서렬특정상사성교고，장저2충작물합병위일류진행분류，병명명위소종작물。결과표명，동소맥、하옥미、춘옥미、면화화소종작물등5류목표가이유효식별，분류총체정도체도90.9%，제도정도분별위94.7%、94.7%、82.4%、86.9%화81.2%，기타미분류유별정도위85.9%。이용중고분변솔요감시간서렬위성영상，재대종농작물시간서렬적변화규률분석기출상，가이준학지제취대종농작물충식면적，재농작물면적자원조사중구유교대적응용잠력。
Remote sensing images with the medium spatial resolution can provide long-time series data of the same area, thus are suitable for remote sensing monitoring of major crops in large scale. Based on the analysis of the time-series spectrum character curve, crop type identification and acreage extraction can be effectively achieved. Taking Hengshui City, Hebei Province as a study area, and employing monthly NDVI (normalized difference vegetation index) time-series data from 16 scenes of HJ-1A/B satellite CCD images with spatial resolution of 30 m, which were collected from October 3rd 2011 to October 24th, 2012, spectrum curve characteristics of the major crop types (winter wheat, summer corn, spring corn, cotton, peanut and soybean) in the whole growth period are extracted. With consideration of high similarity of the NDVI time series among the two crops, i.e. soybean and peanut, they are grouped into the same category to conduct the classification, which is named as minor crop. The NDVI spectrum curve analysis shows that, all other types show a unimodal shape, except for winter wheat/summer corn rotation type; the peaks generally appear in September during the vigorous growth period of crops; consistent with seasonal growth pattern, the NDVI values of both spring corn and cotton during growth period are relatively high, with wider spectrum curve and slow decline; while the spectrum curve of minor crop is relatively narrow, with fast decline. In the study, five parameters, including the NDVI maximum, NDVI minimum, the number of NDVI wave peak, the date of peak and the NDVI value of the most productive period are taken as the extraction characteristics of the five crops and the identification of the five types of crops is conducted in the study area. The precision of the result is evaluated by identifying initial classification threshold, which is gradually adjusted according to the validation of field samples until it is finally confirmed. The distinctive feature for identifying winter wheat/summer corn is its 2 wave peaks. The first date of peak appears between early April and early May and the value of NDVI is above 0.5,and correspondingly, the value of NDVI is below 0.3 in the late March or the early June. The second peak appears between the late August and the middle of September and the value of NDVI is above 0.7, while the value of NDVI is below 0.4 in the early June or the middle of October. With above features, winter wheat/summer corn rotation type can be identified. The number of peak for spring corn is 1, and the peak occurs between late August and the middle of September; the value of NDVI is below 0.6 in the middle of July or the late of September and is above 0.7 in late August or the middle of September; with these features, spring corn can be identified. The number of peak for cotton is 1, and the highest value of NDVI appears between late August and the middle of September; the value of NDVI in the middle of July or late September is above or equal to 0.6 and it is below or equal to 0.5 in early June or the middle of October; according to these features, cotton can be identified. The number of peak for minor crop (soybean and peanut) is 1, and the date of peak appears between late August and the middle of September; the value of NDVI in the middle of July or late September is below 0.6, and is below 0.7 in late August or the middle of September; if having these features, it can be identified as minor crop. By using the decision-tree classification technology based on NDVI, the crop-planting area extraction is carried out. The accuracy of this investigation is verified by on-site GPS measurement of 15 normal example areas with the scale of 2 km × 2 km. The results show that the winter wheat, summer corn, spring corn, cotton and the minor crop can be effectively identified. The general accuracy is as high as 90.9%, and the accuracies for individual crop type are as follows: winter wheat 94.7%, summer corn 94.7%, spring corn 82.4%, cotton 86.9%, minor crop 81.2%, and unidentified crops 85.9%. This paper proves that mass crop’s planting area can be precisely obtained from time-series data of remote sensing images with the medium spatial resolution.