【摘要】：該文針對(duì)農(nóng)業(yè)信息服務(wù)中冬小麥種植面積調(diào)查業(yè)務(wù)的現(xiàn)狀與需求,提出了一種基于NDVI(normal difference vegetation index)時(shí)間序列的冬小麥NDVI加權(quán)指數(shù)(WNDVI,weighted NDVI index)影像算法,可在訓(xùn)練樣本、驗(yàn)證樣本選擇的基礎(chǔ)上實(shí)現(xiàn)冬小麥面積的自動(dòng)提取,并以河北省安平縣及周邊地區(qū)2013-2014年度冬小麥面積提取為例,采用GF-1/WFV(wide field view)數(shù)據(jù)進(jìn)行了算法實(shí)現(xiàn)。算法的主要思路是在時(shí)序影像基礎(chǔ)上,通過(guò)冬小麥NDVI加權(quán)指數(shù)影像的構(gòu)建,擴(kuò)大冬小麥地類(lèi)與其他地類(lèi)的差異,結(jié)合自適應(yīng)的閾值獲取方法,區(qū)分冬小麥地類(lèi),獲取冬小麥作物面積。算法包括冬小麥時(shí)間序列影像的獲取、基于網(wǎng)格的樣本點(diǎn)設(shè)置、構(gòu)建冬小麥NDVI加權(quán)指數(shù)影像、迭代確定冬小麥NDVI加權(quán)指數(shù)提取閾值、精度驗(yàn)證這5個(gè)部分。影像的獲取根據(jù)冬小麥的生長(zhǎng)時(shí)間確定,保證每月1景GF-1/WFV無(wú)云影像,并進(jìn)行預(yù)處理及NDVI計(jì)算;同時(shí)將研究區(qū)劃分為一定數(shù)量的網(wǎng)格,每個(gè)網(wǎng)格再等分為2×2個(gè)子網(wǎng)格,根據(jù)目視解譯、專(zhuān)家知識(shí)、實(shí)地調(diào)查等方法,確定左上網(wǎng)格中心點(diǎn)及右下網(wǎng)格中心點(diǎn)的地物類(lèi)型。統(tǒng)計(jì)該期所有左上網(wǎng)格點(diǎn)冬小麥及其他地物的NDVI均值,冬小麥NDVI大于其他地物的將該期影像的權(quán)值設(shè)置為1,否則設(shè)置為-1,將所有時(shí)相NDVI影像進(jìn)行加權(quán)平均,即可獲取冬小麥NDVI加權(quán)指數(shù)影像。獲取冬小麥NDVI加權(quán)指數(shù)影像后,還需設(shè)置合適的閾值提取冬小麥。該文選用右下網(wǎng)格點(diǎn)目視解譯分類(lèi)結(jié)果作為閾值提取依據(jù),具體方法是將冬小麥指數(shù)從小到大按照一定間隔劃分,作為冬小麥NDVI加權(quán)指數(shù)提取閾值,將各閾值二值法運(yùn)用,與右下網(wǎng)格點(diǎn)的冬小麥提取的目視解譯結(jié)果對(duì)比,精度最高的就是最優(yōu)冬小麥NDVI加權(quán)指數(shù)分割閾值。在所有網(wǎng)格中,以初始識(shí)別獲取的冬小麥面積為準(zhǔn),等概率選擇10個(gè)樣方作為精度驗(yàn)證樣方進(jìn)行驗(yàn)證。精度驗(yàn)證結(jié)果表明分類(lèi)總體精度達(dá)到94.4%,Kappa系數(shù)達(dá)0.88。該文通過(guò)構(gòu)建冬小麥NDVI加權(quán)指數(shù),將比較復(fù)雜的多個(gè)參數(shù)轉(zhuǎn)換為一個(gè)參數(shù),并且農(nóng)學(xué)意義明確,相比傳統(tǒng)的NDVI時(shí)序影像進(jìn)行冬小麥面積的提取,具有自動(dòng)化程度高、面積提取精度高、分類(lèi)結(jié)果穩(wěn)定的特點(diǎn),已經(jīng)在全國(guó)農(nóng)作物面積遙感監(jiān)測(cè)業(yè)務(wù)中進(jìn)行了應(yīng)用。
[Abstract]:In view of the present situation and demand of winter wheat planting area survey in agricultural information service, a NDVI weighted index (WNDVI,weighted NDVI index) image algorithm based on NDVI (normal difference vegetation index) time series was proposed, which can be used in training samples. The automatic extraction of winter wheat area is realized on the basis of sample selection. Taking Anping County of Hebei Province and its surrounding area as an example, the GF-1/WFV (wide field view) data are used to implement the algorithm. The main idea of the algorithm is to expand the difference between winter wheat and other ground types through the construction of winter wheat NDVI weighted index image on the basis of time series image, and to distinguish winter wheat ground type with adaptive threshold acquisition method. To obtain the crop area of winter wheat. The algorithm includes the acquisition of winter wheat time series image, the construction of winter wheat NDVI weighted index image based on the grid sample point setting, the iterative determination of winter wheat NDVI weighted index extraction threshold, and the accuracy verification of the five parts. According to the growth time of winter wheat, the acquisition of image was determined to ensure 1 GF-1/WFV cloud-free image per month, and the preprocessing and NDVI calculation were carried out. At the same time, the study area is divided into a certain number of meshes. Each grid is divided into 2 脳 2 subgrids. According to the methods of visual interpretation, expert knowledge and field investigation, the types of ground objects at the center point of the upper left grid and the center point of the lower right grid are determined. The NDVI mean value of all the left upper grid points of winter wheat and other ground objects in this period is counted. The NDVI of winter wheat is larger than that of other ground objects, and the weight of the image of this period is set to 1, otherwise it is set to -1, and all temporal NDVI images are weighted to average. The NDVI weighted index image of winter wheat can be obtained. After obtaining NDVI weighted index image of winter wheat, it is necessary to set appropriate threshold to extract winter wheat. In this paper, the classification results of visual interpretation of the lower right grid points are selected as the basis of threshold extraction. The specific method is to divide the winter wheat index from small to large according to a certain interval, as the NDVI weighted index of winter wheat to extract the threshold, and apply each threshold binary method. Compared with the visual interpretation results of winter wheat extracted from the lower right grid point, the best NDVI weighted index segmentation threshold is the most accurate. In all meshes, the area of winter wheat obtained by initial identification is taken as the standard, and 10 samples are chosen as precision verification samples to verify the accuracy. The accuracy verification results show that the overall accuracy of the classification is 94. 4% and the Kappa coefficient is 0. 88. In this paper, the NDVI weighted index of winter wheat is constructed, the more complex parameters are converted into one parameter, and the agronomic significance is clear. Compared with the traditional NDVI sequential image, the extraction of winter wheat area has a high degree of automation. The feature of high precision of area extraction and stable classification results has been applied in the field of remote sensing monitoring of crop area in China.
【作者單位】： 中國(guó)農(nóng)業(yè)科學(xué)院農(nóng)業(yè)資源與農(nóng)業(yè)區(qū)劃研究所;
【基金】：“十二五”國(guó)家科技重大專(zhuān)項(xiàng)(高分辨率對(duì)地觀測(cè)系統(tǒng)重大專(zhuān)項(xiàng)“高分農(nóng)業(yè)遙感監(jiān)測(cè)與評(píng)估示范系統(tǒng)(一期)”)
【分類(lèi)號(hào)】：S512.11;S127
，

本文編號(hào)：2300515