【作者】 董芳；

【导师】 赵庚星；

【作者基本信息】 山东农业大学 ， 土壤学， 2012， 博士

【摘要】 遥感技术在农业领域中的应用日益广泛。随着信息技术的发展，越来越多的多尺度空间分辨率数据和多光谱数据的出现为大面积农作物种植面积遥感提取技术提供了海量信息。苹果是我国栽培面积最大、产量最多的水果。山东省作为我国的重要苹果产区之一，苹果栽植面积和产量均居全国前列。对苹果优势区域进行遥感动态监测，掌握苹果园地的面积与分布，对促进我国苹果产业的可持续发展有重要意义。为了准确提取苹果园地信息，本文以栖霞市为研究区，利用不同空间分辨率的多源遥感影像、实测地物光谱数据和GPS调查数据，结合植被指数和DEM信息，采用多种分类方法确定苹果园地面积；并对不同遥感数据源提取苹果园地的适用方法进行了较为系统的研究。论文的研究内容与成果如下：⑴苹果园地遥感识别最佳时相研究利用苹果生长期内的6个时相CBERS影像，分别计算苹果园地、其他果园、耕地的13类植被指数数值，并进行方差分析。结果表明：F检验统计量值最大的月份是四月份，其次是五月份，从而证明利用苹果花期（4月底到5月初）的遥感影像可以有效识别苹果园地。同时，分别从花期的ALOS数据、TM数据、CBERS数据提取苹果园地面积，均取得较好的效果，从而验证了识别苹果园地的最佳时间为苹果花期。⑵ALOS数据花期苹果园地信息提取方法研究研究中，利用BP人工神经网络分别对花期ALOS光谱数据和花期ALOS光谱数据-DEM数据提取苹果园地信息，结果表明：加入DEM信息的人工神经网络分类法提取苹果园地的面积精度较高，空间分辨率中用户精度和生产者精度均为89%以上。证明在提取丘陵区苹果园地信息时，DEM数据是一种必不可少的地理数据。⑶CBERS数据苹果园地信息提取方法研究利用植被指数对花期CBERS影像和多时相CBERS影像进行苹果园地信息提取。结果表明：对花期CBERS影像采用七种植被指数与波段比值指数进行苹果园地提取时，RVI-BAND1/BAND2方法的面积精度和空间精度均最高，其次是RDVI-BAND1/BAND2和MSAVI-BAND1/BAND2方法。利用多时相CBERS影像提取苹果园地时，PVI-SARVI方法在空间精度上明显高于RVI方法。⑷T M数据花期苹果园地信息提取方法研究采用决策树分类法和混合像元分解法提取花期TM影像中的苹果园地。混合像元分解中将实测光谱数据作为分解端元，并利用小波变换对线性分解模型进行改进，采用实测端元改进后线性分解模型、实测端元线性分解模型、TM影像端元线性分解模型分别提取研究区苹果园地信息。结果表明：对比不同信息提取方法发现，利用实测数据作为端元的改进后混合像元分解方法获取的苹果园地面积与统计面积相近，面积提取精度最高，对丰度图像的NDVI值与ALOS数据的平均NDVI值进行回归分析，R2大于0.81，能较好地反映苹果园地的分布。更多还原

【Abstract】 Remote sensing technology has been widely used in agriculture. With the development ofinformation technology, massive data was provided for crop planting area extraction by more andmore multi-resolution and multi-spectral remote sensing image.There is large cultivated area of apple in China, and Shandong Province is one of the mainplangting areas. Dynamic monitoring of dominant apple cultivated area and acquisiton of orchardarea and distribution is significant to Chinese apple industry sustainable development.Taking Qixia City as the research region, using different spatial resolution RS image,measured spectral data and GPS survey data, combined with the NDVI and DEM information, thispaper determined the apple orchard distribution range by the various kinds of classification methodsand conducted a systemic study on appropriate extraction methods for different remote sensing data.The main contents and conclusions are following:⑴Optimal temporal selection for apple orchard classificationBased on six CBERS images in apple growth season, the13kinds of vegetation index of appleorchard, other orchards and cultivated land were calculated, and then, analysis of variance was done.The results showed that F test statistic value in April and May was higher than other periods’. Itproved that the remote sensing images of apple florescence can be used to effectively identify theapple orchard in theory. At the same time, the extraction accuracy from ALOS data, CBERS dataand TM data in apple florescence was satisfying.⑵Apple orchard information extraction method using ALOS dataIn the paper, BP artificial neural network was used to extract apple orchards for ALOSspectrum and ALOS spectrum with DEM data. It showed that BP neural network classifier based onALOS spectrum and DEM data was prior, which the area precision was better, the user accuracyand production accuracy were higher89%. It proved that DEM data was an essential geograhic datafor extraction of apple orchard.⑶Apple orchard information extraction method using CBERS dataIn this section, the apple florescence CBERS and multi-temporal CBERS were used as datasource, and vegetation index was adopted for the apple orchard extraction. Conclusions arefollowings:Comparisons of seven kinds of vegetation index and band ratio index showed that the area andspatial precision of RVI-BAND1/BAND2were the best respectively using the apple florescenceCBERS, followed by RDVI-BAND1/BAND2and MSAVI-BAND1/BAND2. Comparisons of spatial estimation with different vegetation indexes indicated that PVI-SARVI was prior to RVIbased on multi-temporal CBERS.⑷Apple orchard information extraction method using TM dataThis paper determined the apple orchard distribution by the decision tree classification methodand the linear spectral unmixing model. Based on measured spectral endmembers, the WaveletTransform was used to improve linear unmixing models. Three spectral mixture analysis methodsincluding improved linear spectral unmixing model based on measured data, linear spectralunmixing model based on measured data, and linear spectral unmixing model based on TM datawere employed to extract the apple orchard information. The results showde that: after accurateatmospheric and topographic correction, the apple orchard information can be effectively extractedby using the improved linear spectral mixture model based on measured data, and the area precisionwas best; the correlation between NDVI of abundance image and average NDVI of ALOS data wasbetter, with R2higher than0.81.更多还原