本文以冬小麦为测量目标，提出了以中分辨率（TM）数据为核心数据源，低分辨率（MODIS 250米）作为辅助资料和高分辨率（Quickbird 2.4米）作为验证资料源，以对地抽样理论为依托的冬小麦测量体系。该测量体系中，在对地统计抽样模型的基础上，估算整体监测区内冬小麦的总量，完成总量控制下的中分辨率冬小麦面积信息提取；中分辨率测量结果为低分辨率遥感数据混合像元分解提供样本，进行中低分辨率一致性测量；低分辨率时间序列的测量结果可以进一步修正中分辨率由于同物异谱、异物同谱造成的冬小麦错出与错入现象，测量结果以高分辨率和野外调查数据作为真实数据作为验证。得到主要结论如下：（1）以冬小麦模拟图像作为真值，在对地抽样理论支持下，采用随机抽样、系统抽样、分层抽样抽样三种抽样方式，抽样框大小定义为10×10，抽样样本量为1%的前提下，采用三种抽样方法进行抽样反推冬小麦种植面积总量。结果表明，三种抽样方法可以达到99%以上抽样精度。从抽样精度的方差、极差两个评价指标可以看出，分层抽样精度的方差、极差都比较小，证明分层抽样能够保证抽样结果的稳定性。因此，分层抽样是三种抽样方式中最为有效的抽样方法，适合于冬小麦种植面积的抽样测量。（2）以冬小麦模拟图像作为真值，在95%置信度95%精度的前提下，分层层数对抽样精度、抽样比和抽样方差的影响比较大。在不同抽样格网尺寸下，冬小麦抽样精度都比较高且稳定，保持在98%以上。以冬小麦自身进行分层抽样反推自身总量，随着分层层数的增加，抽样比、抽样方差降低比较快，为了降低样本测量工作量，增加分层层数可以降低样本量。（3）针对作物区分别布置10%，20%，30%，40%，50%，60%，70%，80%，90%，100%随机冬小麦样本。以作物区作为分层指针，在最优抽样样本量的前提下（统计学上的抽样精度），选用8×8抽样框进行抽样，最终反推冬小麦的抽样精度。在同一冬小麦丰度的前提下，随着层数的增加，冬小麦的抽样精度降低，呈发散趋势；随着作物区内的冬小麦覆盖丰度升高，冬小麦的抽样精度随着层数的增加，抽样精度降低比较慢。在同一分层时，冬小麦的丰度越高，抽样精度也就越高。其中，当作物区内冬小麦丰度为60%的时候，冬小麦的抽样精度都非常高且稳定。（4）在作物区内60%冬小麦的前提下，在0.1，0.01，0.005，0.001，0.0005，0.0001固定抽样样本量的前提下，以作物区作为分层指标进行2-15层分层抽样。在同一抽样量的前提下，抽样反推出抽样精度随着层数出现先增高后降低的趋势，抽样方差也是随着分层的增加先增高后降低，抽样极差正好相反，出现先降低后增高的趋势。而且，随着抽样量的降低，这种变化趋势更为明显，说明了适当的进行分层是可以提高抽样精度的。在0.5%抽样量来看，分层数量定义为8层分时，抽样精度比较高，抽样的极差也比较小。（5）模拟图像方法验证了分层抽样方法能够有效地提取出整区的冬小麦种植面积。通过分层抽样方法，对TM提取出的作物区进行抽样，以Quickbird数据提取出的冬小麦作为真值，在完备Quickbird支撑抽样方法下，能够有效地反推出整个TM内的冬小麦种植面积，平均精度在97%以上，最低抽样误差也高于95%。在不完备Quickbird数据支撑TM区抽样下，用Quickbird区的样本来替代非Quickbird覆盖下TM区的样本，这样的替代方法是比较有效的，能够保证比较高的抽样精度，这样就解决了在只有一小部分Quickbird数据覆盖的前提下，采用样本替代的方法能够较少Quickbird数据的购置，从而降低了测量成本，又可以保证抽样精度。通过对分区抽样图像的目视解译，可以获得准确的冬小麦的种植面积。（6）TM图像在总量控制的前提下，综合了软硬分类法各自的优点，采用软分类（SVR混合像元分解）和硬分类（模拟识别方法）相结合的方法进行了冬小麦信息提取，建立总量控制下的中分辨率冬小麦测量模型，该方法测量的提取出的冬小麦空间位置精度（86.8%）要高于单纯的软分类方法（线性分解方法）和硬分类方法（ISODATA），能够有效地剔除道路边缘或者一定非冬小麦的区域，保证冬小麦空间分布的准确性。（7）在TM与MODIS的共区范围内MODIS的SVR混合像元分解结果的区域精度、像元精度随着输入样本量的增加而增高。当样本量达到10%的时候，全区的区域精度达到98%以上，像元精度达到93%以上；冬小麦区的区域精度达到98%以上，像元精度达到92%以上；非冬小麦区的区域精度和像元精度相等都达到96%以上，而且都比较稳定。因此，10%的TM样本量是可以满足MODIS冬小麦的混合像元分解，准确地提取出冬小麦面积。TM区冬小麦的样本质量的误差要低于50%，以保证MODIS混合像元分解的结果的区域精度、像元精度在89%，86%以上。 （8）通过TM区提取给非TM区提供样本进行非TM区MODIS混合像元分解的研究表明，非TM区MODIS混合像元结果的区域精度、像元精度随着输入样本数量的增加而增高，在20%样本量下，区域精度和像元精度稳定下来，分别在97%和92%以上。基于这一测量结果的一致性，利用MODIS数据，在TM区域内提取样本可以准确地提取出非TM区内冬小麦的种植面积。因此，TM数据的支持下，利用MODIS数据进行大范围的遥感测量冬小麦面积，解决无法获取最佳时相中分辨率图像冬小麦提取的数据问题。（9）由于单期TM数据记录瞬间的地物状况，存在同物异谱、异物同谱的现象，导致冬小麦提取中发生错入、错出。MODIS能够有效地反映出地物的时间序列特征，而且MODIS混合像元分解能够有效地提取出冬小麦的种植面积，与TM测量结果具有很强的一致性，因此可以利用MODIS测量结果对TM测量结果进行修正。本研究表明，通过测量结果的分析，对TM冬小麦测量结果的错入、错出区域能够有效地辨别出来。错出区域主要是光谱信息与典型的冬小麦相异，但是在MODIS数据时间序列曲线上反映出明显的冬小麦光谱特征，在本论文的研究中证实这是可行的。综上，在对地抽样的基础上，对中尺度冬小麦面积测量进行控制；中、低分辨率冬小麦的一致性测量，保证了大范围冬小麦测量结果，同时解决了中分辨率同物异谱、异物同谱现象，所形成的对地抽样多尺度遥感资料复合冬小麦测量体系切实实用的。

In this paper, aiming at the wheat area extraction, the wheat area survey system based spatial sample theory was developed, in which the medium resolution image (TM) is the key data, and the low resolution (MODIS) and high resolution image (Quickbird) are the assistant and validated data. The spatial sample model is used for evaluating the wheat area in the whole region, which is taken for the medium image extraction. And samples from the result of the medium image are for the low resolution image mixed spectral unmixing. Finally, the wheat result from the low resolution can correct the sample thing with different spectrum and the different thing with same spectrum which happen in the medium resolution image. Some main conclusions can be drawn as followes:Based on the spatial theory and 1% sample amount, the random, systematic and stratified methods are adopted for calculating the whole regional area, the stratified method is the best method among three sample method, which can obtain the best and stable accuracy.The strata number influences the sample accuracy, ratio and variance. With the different sample size, the crop sample accuracies are high and stable, and sample accuracy variances are low. The sample ratio, sample variance decreases with the strata layer increasement. With the different wheat ratio within the crop region and the optimum sample amount, the stratified sample method is taken for wheat extraction. The wheat area accuracy decreases with the layer number increasement. While with the wheat ratio is higher, the sample accuracy increases slowly. When the 60% wheat is within the crop region, the wheat sample accuracy is high and stable.With 60% wheat area within the crop region and the fixed sample amount set, the wheat sample accuracy goes up first and then down, and the sample variance is just like this. So the reasonable layer amount can improve the sample accuracy. The experiment shows that the 0.1% sample and eight layer amount is suitable.The wheat from the Quickbird as the true value, the TM wheat area can be obtained by the stratified sample method with the completed Quickbird assistant. The average accuracy is above 97%. The sample can be instead from the Quickbird region when there is not enough Quickbird for sample, which can also achieve the enough aaccuracy. Finally, with visiual sample imag interpretation, the accurate wheat area can also be obtained. All of above experiment can help wheat area extraction with the actual remotely sensing data.The method of simulating image proves that stratify sampling method can extract the area of winter wheat in the entire test district effectively. By stratify sampling, we can reason out the area of winter wheat in the TM image above 95% of the precision, when the lowest precision isn’t higher than 95%, with the data of sampling in TM, the abstract data of the Quickbird Image which is think to be the true value and in the condition of sampling by Quickbird. On the other hand, in the condition that sampling by Quickbird, the way that the sample of TM is replace by that of Quickbird is effective and can confirm higher sampling precision, which is not only reduce the measure cost, but also confirm the sampling precision. Through the observation interpretation to the subarea sampling image, the area of winter wheat can be measured exactly.Incorporating the advantages of soft and hard classification, the combination of soft and hard classification is suitable for the wheat area extraction with the gross wheat area domination, which is the model of the wheat area extraction with middle resolution image. This method is better than the soft classification ( linear spectral unmixing method) and hard classification( ISODATA), which can eliminate pixels from the edge of road and typical non-wheat in order to ensure the accuracy.In the share area of the TM MODIS data, the regional and pixel accuracy of SVR mixed unmixing method for the MODIS data is higher as soon as the input sample amount. When the sample amount is 10%, the regional and pixel accuracy in the whole region are about 98% and 93% respectively. The regional and pixel accuracy in the wheat region are 98% and 92% respectively. The regional and pixel accuracy in the non-wheat region are about 96%. All above accuracies are stable. So 10% sample amount is enough for the MODIS wheat area extraction with the SVR method. The sample quality influences the MODIS mixed spectral unmixing method. The error of TM sample should be lower than 50%, which can ensure the regional and pixel accuracy above the 89% and 86%.According the MODIS without TM for mixed pixels unmixing, the pixel accuracy and regional accuracy increase with the sample amount increasement. When sample amount is 20%, the regional accuracy and pixel accuracy are stable, about 97% and 92% respectively. According the theory, the optimal wheat result can be obtained from the MODIS data with the TM sample.For the instantaneous recode the land condition with the single date TM image, the same thing with the different spectral and the different thing the same spectral can not be avoided. MODIS data can show the characters of time, which can obtain the wheat plant area with mixed pixel unmixing method effectively. And the result is consistent with result come from the TM data. So the MODIS result can correct the result coming from the TM data. The research shows that the MODIS correction is useful.In all, the accurate wheat area can be obtained from the middle resolution image based on the spatial sample, and the consist survey between the medium image and low resolution image ensure the large scale wheat area extraction, which can also resolve the phenomenon about the same thing the different spectral or the different thing with the same spectral. The wheat survey system is effective for wheat area survey with the completed or non-completed image data.