基于Sentinel-2时序多特征的植被分类

植被分类是研究森林资源状况和动态变化规律的基础,利用遥感手段可以更加快速、准确地识别植被类型。以位于内蒙古赤峰市喀喇沁旗西南部的旺业甸实验林场为研究对象进行植被分类。采用分层分类的思想,首先根据植被物候特征选取植被生长旺盛时期的影像,计算归一化植被指数（normalized difference vegetation index,NDVI）并设定合适的阈值将研究区内的植被提取出来,剩余部分归为非植被。然后选取NDVI时间序列、最佳时相的Sentinel-2数据中10个波段的光谱反射率特征和主成分分析前3个分量的纹理特征作为分类特征,利用支持向量机分类器将研究区内的植被类型分为耕地、草地、常绿针叶林、落叶针叶林和落叶阔叶林五大类,并将分类结果与最大似然法、NDVI时序+光谱特征的分类结果进行对比分析。NDVI时序+光谱特征+纹理特征的多特征植被分类总体精度达87.64%,Kappa系数为0.85,分别比最大似然法和结合NDVI时序+光谱特征的分类总体精度提高了15.73%和14.61%,Kappa系数提高了0.20和0.18。其中常绿针叶林和耕地的分类结果与实地调查情况高度一致,分类精度分别达到95.65%和92.31%。从而得出:①基于多特征的分类方法有助于提高分类精度;②NDVI时序特征对于植被的区分具有很大帮助;③采用分层分类的思想,首先将研究区内的植被提取出来,可以排除非植被因素的干扰,有效提高植被类型的分类精度。     

Leaf chlorophyll content retrieval of wheat by simulated RapidEye,Sentinel-2 and EnMAP data

Leaf chlorophyll content(LCC) is an important physiological indicator of the actual health status of individual plants. An accurate estimation of LCC can therefore provide valuable information for precision field management. Red-edge information from hyperspectral data has been widely used to estimate crop LCC. However, after the advent of red-edge bands in satellite imagery, no systematic evaluation of the performance of satellite data has been conducted. Toward this end, we analyze herein the performance of winter wheat LCC retrieval of currant and forthcoming satellites(RapidEye, Sentinel-2 and EnMAP) and their new red-edge bands by using partial least squares regression(PLSR) and a vegetation-indexbased approach. These satellite spectral data were obtained by resampling ground-measured hyperspectral data under various field conditions and according to specific spectral response functions and spectral resolution. The results showed: 1) This study confirmed that RapidEye, Sentinel-2 and EnMAP data are suitable for winter wheat LCC retrieval. For the PLSR approach, Sentinel-2 data provided more accurate estimates of LCC(R2=0.755, 0.844, 0.805 for 2002, 2010, and 2002+2010) than do RapidEye data(R2=0.689, 0.710, 0.707 for 2002, 2010, and 2002+2010) and EnMAP data(R2=0.735, 0.867, 0.771 for 2002, 2010, and 2002+2010). For index-based approaches, the MERIS terrestrial chlorophyll index, which is a vegetation index with two red-edge bands, was the most sensitive and robust index for LCC for both the Sentinel-2 and EnMAP data(R2≥0.628), and the indices(NDRE1, SRRE1 and CIRE1) with a single red-edge band were the most sensitive and robust indices for the RapidEye data(R2≥0.420); 2) According to the analysis of the effect of the wavelength and number of used red-edge spectral bands on LCC retrieval, the short-wavelength red-edge bands(from 699 to 734 nm) provided more accurate predictions when using the PLSR approach, whereas the long-wavelength red-edge bands(740 to 783 nm) gave more accurate predictions when using the vegetation indice(VI) approach. In addition, the prediction accuracy of RapidEye, Sentinel-2 and EnMAP data was improved gradually because of more number of red-edge bands and higher spectral resolution; VI regression models that contain a single or multiple red-edge bands provided more accurate predictions of LCC than those without red-edge bands, but for normalized difference vegetation index(NDVI)-, simple ratio(SR)-and chlorophyll index(CI)-like index, two red-edge bands index didn't significantly improve the predictive accuracy of LCC than those indices with a single red-edge band. Although satellite data with higher spectral resolution and a greater number of red-edge bands marginally improve the accuracy of estimates of crop LCC, the level of this improvement remains insufficient because of higher spectral resolution, which results in a worse signal-to-noise ratio. The results of this study are helpful to accurately monitor LCC of winter wheat in large-area and provide some valuable advice for design of red-edge spectral bands of satellite sensor in future.     

Separability of coffee leaf rust infection levels with machine learning methods at Sentinel-2 MSI spectral resolutions

Coffee leaf rust (CLR) caused by the fungus Hemileia vastarix is a devastating disease in almost all coffee producing countries and remote sensing approaches have the potential to monitor the disease. This study evaluated the potential of Sentinel-2 band settings for discriminating CLR infection levels at leaf levels. Field spectra were resampled to the band settings of the Sentinel-2, and evaluated using the random forest (RF) and partial least squares discriminant analysis (PLS-DA) algorithms with and without variable optimization. Using all variables, Sentinel-2 Multispectral Imager (MSI)-derived vegetation indices achieved higher overall accuracy of 76.2% when compared to 69.8% obtained using raw spectral bands. Using the RF out-of-bag (OOB) scores, 4 spectral bands and 7 vegetation indices were identified as important variables in CLR discrimination. Using the PLS-DA Variable Importance in Projection (VIP) score, 3 Sentinel-2 spectral bands (B4, B6 and B5) and 5 vegetation indices were found to be important variables. Use of the identified variables improved the CLR discrimination accuracies to 79.4 and 82.5% for spectral bands and indices respectively when discriminated with the RF. Discrimination accuracy slightly increased through variable optimization for PLS-DA using spectral bands (63.5%) and vegetation indices (71.4%). Overall, this study showed the potential of the Sentinel 2 MSI band settings for CLR discrimination as part of crop condition assessment. Nevertheless further studies are required under field conditions.     

Strategies for monitoring within-field soybean yield using Sentinel-2 Vis-NIR-SWIR spectral bands and machine learning regression methods

Soybean crop plays an important role in world food production and food security, and agricultural production should be increased accordingly to meet the global food demand. Satellite remote sensing data is considered a promising proxy for monitoring and predicting yield. This research aimed to evaluate strategies for monitoring within-field soybean yield using Sentinel-2 visible, near-infrared and shortwave infrared (Vis/NIR/SWIR) spectral bands and partial least squares regression (PLSR) and support vector regression (SVR) methods. Soybean yield maps (over 500 ha) were recorded by a combine harvester with a yield monitor in 15 fields (3 farms) in Paraná State, southern Brazil. Sentinel-2 images (spectral bands and 8 vegetation indices) across a cropping season were correlated to soybean yield. Information pooled across the cropping season presented better results compared to single images, with best performance of Vis/NIR/SWIR spectral bands under PLSR and SVR. At the grain filling stage, field-, farm- and global-based models were evaluated and presented similar trends compared to leaf-based hyperspectral reflectance collected at the Brazilian National Soybean Research Center. SVR outperformed PLSR, with a strong correlation between observed and predicted yield. For within-field soybean yield mapping, field-based SVR models (developed individually for each field) presented the highest accuracies. The results obtained demonstrate the possibility of developing within-field yield prediction models using Sentinel-2 Vis/NIR/SWIR bands through machine learning methods.

     

多时相Sentinel-2影像在浙西北茶园信息提取中的应用

利用Sentinel-2遥感影像研究一种快速、准确提取茶园空间分布的新方法,可为茶园经济林资源及其动态变化的快速检测提供新的手段。以浙江省西北部为研究区,根据实地调查选取6类典型植被,基于4个季节的Sentinel多光谱影像分析不同植被物候及光谱特征。茶园在5月经历修剪后与其他植被区别较大,根据红边与短波红外波段构建归一化茶园指数（NDTI）。基于新指数建立决策树模型提取茶园,通过谷歌地球对结果进行验证。结果显示:归一化茶园指数可以最大限度扩大茶园与其他植被之间的差距。基于该指数提取茶园的总精度达93.83%,Kappa系数为0.917,成功实现了浙西北茶园信息的提取,证明了使用红边波段提取茶园的潜力。     

基于Sentinel-1/2数据的中国南方单双季稻识别结果一致性分析

【目的】微波遥感因具有全天时、全天候数据获取的特点,在多云雨的中国南方水稻识别研究中表现出巨大潜力。本研究通过对比Sentinel-1SAR遥感数据和Sentinel-2光学遥感数据用于水稻遥感制图的效果,分析光学和SAR遥感数据对于单双季稻识别结果的一致性,并探索水稻识别的最优SAR影像特征。【方法】本研究使用Sentinel-1/2卫星数据,基于面向对象的随机森林分类算法和Google Earth Engine平台,提取洞庭湖平原4个典型水稻种植区的单双季稻空间分布。通过比较9种不同传感器和特征组合场景的分类精度和分类结果统计指标,并计算NDVI和SAR特征时序（VH、VV、VH/VV）的R²和DTW距离,分析识别单双季稻的最优SAR特征,评估光学和SAR遥感数据对于单双季稻识别结果的一致性。【结果】VH、VV和VH/VV时序识别单双季的总体精度分别为90.42%、82.08%和88.33%,而联合VH和VH/VV时序识别单双季稻的总体精度可达91.67%。VH（VH/VV、VV）时序与单双季稻NDVI时序的R²和DTW距离分别为0.870（0.915、0.986）、4.715（1.896、5.506）（单季稻）和0.597（0.783、0.673）、2.396（1.839、3.441）（双季稻）。较高的R²和较低的DTW距离说明单双季稻的VH/VV时序与NDVI时序相关度更高,可以较好地反映单双季稻的生长周期规律。同时,VH可以较好地反映单双季稻移栽期的淹水特征。基于光学数据和SAR数据在6个时间窗口的特征（S-2：NDVI、EVI、LSWI;S-1：VH、VH/VV）识别单双季稻的总体精度分别为91.25%和90.00%,识别结果面积相关性可达95.70%。【结论】SAR遥感数据与光学遥感数据水稻识别结果一致性较高。应用Sentinel-1在多云雨区识别单双季稻具有巨大潜力,VH和VH/VV后向散射系数时序是识别水稻的优质特征。研究结果为多云多雨区使用SAR数据进行特征优选以高精度识别单双季稻提供了重要技术支撑。     

Mapping winter rapeseed in South China using Sentinel-2 data based on a novel separability index

Large-scale crop mapping using remote sensing data is of great significance for agricultural production, food security and the sustainable development of human societies. Winter rapeseed is an important oil crop in China that is mainly distributed in the Yangtze River Valley. Traditional winter rapeseed mapping practices are insufficient since they only use the spectral characteristics during the critical phenological period of winter rapeseed, which are usually limited to a small region and cannot meet the needs of large-scale applications. In this study, a novel phenology-based winter rapeseed index (PWRI) was proposed to map winter rapeseed in the Yangtze River Valley. PWRI expands the date window for distinguishing winter rapeseed and winter wheat, and it has good separability throughout the flowering period of winter rapeseed. PWRI also improves the separability of winter rapeseed and winter wheat, which traditionally have been two easily confused winter crops. A PWRI-based method was applied to the Middle Reaches of the Yangtze River Valley to map winter rapeseed on the Google Earth Engine platform. Time series composited Sentinel-2 data were used to map winter rapeseed with 10 m resolution. The mapping achieved a good result with overall accuracy and kappa coefficients exceeding 92% and 0.85, respectively. The PWRI-based method provides a new solution for high spatial resolution winter rapeseed mapping at a large scale.     

Early-season crop type mapping using 30-m reference time series

Early-season crop type mapping could provide important information for crop growth monitoring and yield prediction, but the lack of ground-surveyed training samples is the main challenge for crop type identification. Although reference time series based method(RBM) has been proposed to identify crop types without the use of ground-surveyed training samples, the methods are not suitable for study regions with small field size because the reference time series are mainly generated using data set with low spatial resolution. As the combination of Landsat data and Sentinel-2 data could increase the temporal resolution of 30-m image time series, we improved the RBM by generating reference normalized difference vegetation index(NDVI)/enhanced vegetation index(EVI) time series at 30-m resolution(30-m RBM) using both Landsat and Sentinel-2 data, then tried to estimate the potential of the reference NDVI/EVI time series for crop identification at early season. As a test case, we tried to use the 30-m RBM to identify major crop types in Hengshui, China at early season of 2018, the results showed that when the time series of the entire growing season were used for classification, overall classification accuracies of the 30-m RBM were higher than 95%, which were similar to the accuracies acquired using the ground-surveyed training samples. In addition, cotton, spring maize and summer maize distribution could be accurately generated 8, 6 and 8 weeks before their harvest using the 30-m RBM; but winter wheat can only be accurately identified around the harvest time phase. Finally, NDVI outperformed EVI for crop type classification as NDVI had better separability for distinguishing crops at the green-up time phases. Comparing with the previous RBM, advantage of 30-m RBM is that the method could use the samples of the small fields to generate reference time series and process image time series with missing value for early-season crop classification; while, samples collected from multiple years should be further used so that the reference time series could contain more crop growth conditions.     

Comparison of machine learning algorithms for mapping mango plantations based on Gaofen-1 imagery

Mango is a commercial crop on Hainan Island, China, that is cultivated to develop the tropical rural economy. The development of accurate and up-to-date maps of the spatial distribution of mango plantations is necessary for agricultural monitoring and decision management by the local government. Pixel-based and object-oriented image analysis methods for mapping mango plantations were compared using two machine learning algorithms (support vector machine (SVM) and Random Forest (RF)) based on Chinese high-resolution Gaofen-1 (GF-1) imagery in parts of Hainan Island. To assess the importance of different features on classification accuracy, a combined layer of four original bands, 32 gray-level co-occurrence (GLCM) texture indices, and 10 vegetation indices were used as input features. Then five different sets of variables (5, 10, 20, and 30 input variables and all 46 variables) were classified with the two machine learning algorithms at object-based level. Results of the feature optimization suggested that homogeneity and variance were very important variables for distinguishing mango plantations patches. The object-based classifiers could significantly improve overall accuracy between 2–7% when compared to pixel-based classifiers. When there were 5 and 10 input variables, SVM showed higher classification accuracy than RF, and when the input variables exceeded 20, RF showed better performances. After the accuracy achieved saturation points, there were only slightly classification accuracy improvements along with the numbers of feature increases for both of SVM and RF classifiers. The results indicated that GF-1 imagery can be successfully applied to mango plantation mapping in tropical regions, which would provide a useful framework for accurate tropical agriculture land management.     

联合光学和合成孔径雷达数据的太平湖森林地上生物量反演研究

光学和合成孔径雷达（SAR）多源传感器数据融合对提高森林地上生物量（AGB）提取精度具有重要意义。以太平湖森林为研究对象,以Sentinel-1 SAR数据和Sentinel-2光学数据为数据源,利用随机森林回归算法系统性地评估光学和SAR数据对AGB反演的互补优势和策略选择。采用Sentinel-2光学数据的AGB反演精度（R²=0.63,RMSE=37.05 mg/hm²,sMAPE=0.56）优于采用Sentinel-1 SAR数据的AGB反演精度（R²=0.37,RMSE=52.25 mg/hm²,sMAPE=0.65）,联合两者数据的AGB估算精度最高（R²=0.69,RMSE=34.17 mg/hm²,sMAPE=0.55）;基于不同策略构建的AGB估计模型当中,植被指数（RVI、NDVI和红边相关的NDVIre）和纹理变量（NDVIre_Mea）的重要性高于光谱波段和后向散射系数。联合Sentinel-1和Sentinel-2数据的光谱波段、植被指数、纹理信息和后向散射系数,能够有效的缓和遥感信息饱和性问题和提高AGB反演精度。     

应用哨兵2A多时相遥感影像对树种的识别

森林树种类别的准确识别和专题图制作对于森林经营管理具有重要作用。以哨兵2A多时相遥感影像数据源,通过对单时相影像以及根据基尼系数(Gini系数)和信息熵进行的不同影像组合,利用随机森林算法对大孤家林场的树种类别进行识别和分类制图。结果表明:哨兵2A多时相影像组合能够显著改善树种分类的精度,多时相影像组合总体分度最高为87.45%,Kappa系数为0.825,比单时相遥感影像分类总体精度提升了4%~7%。     

基于高空间分辨率卫星影像的新疆阿拉尔市棉花与枣树分类

【目的】 枣树和棉花是新疆地区的两大优势作物。利用高空间分辨率遥感影像对作物进行识别,更加快速、准确地获取枣树和棉花的种植面积及其分布区域,以利于相关部门政策的制定及农作物的精确管理。【方法】 本文以新疆阿拉尔市主要农作物为研究对象,运用基于像素与面向对象的遥感影像分类方法,通过比较光谱角制图（SAM）、支持向量机（SVM）、CART决策树（DTs）、随机森林（RF）这4种机器学习算法在高空间分辨率卫星影像分类中的作物识别精度,探究影像获取时期（2016-05-10、2016-09-07、2016-10-08）及面向对象的信息提取技术对作物分类精度的影响。【结果】 5月份影像（即棉花覆膜期影像）作物分类精度最高,10月份影像次之,9月份影像最差;与基于像素的作物分类方法相比,面向对象的作物分类方法可以使各时期的作物分类总体精度得到一定提高（除SAM之外）,各时期分类精度分别提高了4.83%、7.77%、7.22%,最高分类精度分别为93.52%（2016-05-10）、85.36%（2016-09-07）、88.88%（2016-10-08）,均实现了较好的作物分类效果。【结论】 5月份（棉花覆膜期）影像对棉花和枣树分类效果最好,该时期的棉花被地膜覆盖,且枣树表现出明显的植被光谱特性,两种作物生长早期呈现出差异化的光谱特征,因此棉花和枣树的遥感识别应在作物生长早期进行;面向对象的分类方法可以综合运用光谱、纹理及空间信息,特别是纹理信息的加入,可以取得比基于像素方法更高的分类精度,且提供一种高效提取田块边界的手段,对当地农田信息化管理具有重要应用价值。在棉花和枣树识别过程中,纹理特征的重要性高于光谱和空间特征,红光和绿光波段在所有波段中对棉花和枣树的识别贡献最大。     

Winter wheat yield estimation based on assimilated Sentinel-2 images with the CERES-Wheat model

Assimilating Sentinel-2 images with the CERES-Wheat model can improve the precision of winter wheat yield estimates at a regional scale. To verify this method, we applied the ensemble Kalman filter(EnKF) to assimilate the leaf area index(LAI) derived from Sentinel-2 data and simulated by the CERES-Wheat model. From this, we obtained the assimilated daily LAI during the growth stage of winter wheat across three counties located in the southeast of the Loess Plateau in China: Xiangfen, Xinjiang, and Wenxi. We assigned LAI weights at different growth stages by comparing the improved analytic hierarchy method, the entropy method, and the normalized combination weighting method, and constructed a yield estimation model with the measurements to accurately estimate the yield of winter wheat. We found that the changes of assimilated LAI during the growth stage of winter wheat strongly agreed with the simulated LAI. With the correction of the derived LAI from the Sentinel-2 images, the LAI from the green-up stage to the heading–filling stage was enhanced, while the LAI decrease from the milking stage was slowed down, which was more in line with the actual changes of LAI for winter wheat. We also compared the simulated and derived LAI and found the assimilated LAI had reduced the root mean square error(RMSE) by 0.43 and 0.29 m2 m–2, respectively, based on the measured LAI. The assimilation improved the estimation accuracy of the LAI time series. The highest determination coefficient(R2) was 0.8627 and the lowest RMSE was 472.92 kg ha–1 in the regression of the yields estimated by the normalized weighted assimilated LAI method and measurements. The relative error of the estimated yield of winter wheat in the study counties was less than 1%, suggesting that Sentinel-2 data with high spatial-temporal resolution can be assimilated with the CERES-Wheat model to obtain more accurate regional yield estimates.     

一年一季农作物遥感分类的时效性分析

【目的】基于遥感影像的作物分类研究是提取作物种植面积和长势分析及产量估测的基础,也是推动现代化农业快速发展的动力。研究结果可为农业等相关部门掌握农情,进行宏观调控提供依据。目前,农业遥感研究主要集中于中低分辨率遥感影像,影响植被信息提取的精度,应用高分辨率多时相遥感影像和选择最优分类方法可以提高植被信息提取精度。明确农作物遥感分类的时效性与最优分类方法,为快速、准确地获取作物空间分布数据和农情定量遥感监测提供依据。【方法】基于黑龙江省虎林市2014年5—10月覆盖完整生长期的20幅遥感影像,构建16 m分辨率NDVI时间序列曲线,建立决策树分类模型,通过分类影像进行系列阈值分割,并结合辅助背景数据及专家知识,成功提取虎林市土地利用覆被信息;利用20幅影像依次波段合成的方式进行作物分类,明确最优时相;将提取的耕地范围作为作物分类规则,并与未提取耕地范围的作物分类结果进行比较;同时通过最大似然法、马氏距离法、神经网络法、最小距离法、支持向量机、波谱角分类法、主成分分析法多种分类方法进行作物分类;利用农业保险投保地块数据进行精度验证。【结果】(1)7月初、7月末到8月初、9月末是研究区一年一季作物遥感分类的3个关键时相;(2)决策树分类方法在提取土地利用覆被信息的结果中精度最高,总体精度90.24%,Kappa系数0.87;(3)6月初与7月初2幅影像结合采用最大似然法对作物进行分类的总体精度高达94.01%,Kappa系数为0.79,6月初与7月初的影像结合,可以解决作物分类的时效性;(4)结合9月21日的影像,总体精度进一步提高,大豆分类精度明显提高,最终确定最大似然法为最优作物分类方法。【结论】通过遥感数据能实现在7月上旬对作物进行精准分类,拓展了遥感数据在农业领域的应用价值,对一年一季地区作物快速分类与农情定量遥感监测有重要意义。     

Sentinel-2A多特征变量反演针叶林地上生物量能力评估

  目的  森林生物量是衡量森林碳储量的关键因子，准确估算生物量对掌握森林现状和森林资源合理利用具有重要意义。欧空局发射Sentinel-2A数据因其丰富的光谱信息和较高的空间分辨率为生物量的反演和监测提供了新的机会。本文旨在评估基于Sentinel-2A的各类特征变量反演针叶林地上生物量的能力以及完成区域尺度的针叶林地上生物量定量估测。  方法  试验以内蒙古赤峰市喀喇沁旗旺业甸林场针叶林为研究对象，以Sentinel-2A为主要数据源，提取了10个波段反射率、20个植被指数和5个生物物理参数共3种类型变量，分别建立基于光谱反射率、植被指数、生物物理参数，以及融合3类变量的多元逐步回归生物量估算模型，同时每组均加入高程因子分析地形对估算精度的影响。  结果  （1）基于多种类型参数建立的模型估算效果最好，模型决定系数达到0.765，均方根误差为39.49 t/hm2；（2）在3组单类型变量模型中，基于植被指数的预测结果最好，说明相比于波段反射率和生物物理参数，植被指数对针叶林地上生物量的估算贡献更大；（3）无论基于何种类型参数建模，高程信息的加入都会提高针叶林地上生物量的估算精度。  结论  基于Sentinel-2A植被指数与地形特征的针叶林地上生物量反演模型较好，可用于区域生物量估算。该研究对区域性森林资源监测的实际应用具有指导意义。      

基于遥感影像的土地利用特征提取与城乡梯度差异分析——以河北省涿州市为例

为快速获取区域土地利用特征和精细刻画城乡土地利用差异,以河北省涿州市为研究对象,基于Sentinel-2影像数据,采取面向对象方法进行影像分割,利用隶属度函数与决策树方法相结合的非监督分类算法对涿州市土地利用进行分类,并选取了不同方向的城乡梯度样带进行了土地利用特征分析。结果表明,应用模糊决策树方法的涿州市土地利用分类结果总体精度为93.7%,Kappa系数0.892,分类精度较高。分析上述结果发现:涿州市土地利用类型以耕地与城乡居民点用地为主,林地、草地、水体等自然生态空间比例较低,土地利用的城乡梯度特征明显;耕地集中分布在距离城市中心4～7km的东南、南、西方向;城乡居民点整体分布分散,在距离城市中心3km以内、5km、8～9km呈现明显的集聚特征。建议涿州市依据预期人口规模和集聚特征优化建设用地布局,提高建设用地集约利用强度,同时提高林地、草地、水体等生态空间比例。     

基于水稻特征波段的决策树分类研究

以滁州市为例,结合水稻物候的特征波段,选用反映水稻物候期时相的TM数据,并基于多特征波段,构建CART决策树分类提取水稻种植面积。结果表明,植被指数、湿度因子、绿度因子、纹理特征等多特征参与CART决策树分类能够提高总体精度。基于光谱信息、植被指数和纹理特征的决策树分类的总精度比以最大似然法进行的监督分类方法提高了6.942 1百分点,Kappa系数提高了0.110 4。合理选用作物物候期数据及其遥感影像的特征波段能够有效降低分类误差,为地形复杂地区获取作物种植面积提新的方法。     

基于Sentinel-2多时相影像的果树种植区遥感提取

为提取果树的空间分布信息,以果树生长期内不同月份的Sentinel-2多光谱遥感影像为数据源,以大沙河流域果树为研究对象,通过分析不同月份的光谱信息得出最佳监测时期,并在此基础上,选择不同时期的5种植被指数[归一化植被指数(NDVI)、比值植被指数(RVI)、增强型植被指数(EVI)、结构密集型色素指数(SIPI)和归一化水指数(NDWI)],结合机器学习技术构建决策树提取模型。结果发现,3、4、7、8月份的影像适于果树面积提取。通过Feature_importances_属性筛选出贡献度高的不同时期的植被指数作为输入特征,结合超参数学习曲线和网格搜索技术确定决策树模型的Max_depth和Min_samples_leaf参数分别为5和10时模型的效果最佳。参数调整后绘制决策树模型,模型在训练集和测试集上的精度分别达到了0.919 4和0.875 1。提取结果表明,研究区内的果树主要种植在大沙河两岸,东部与西北部的果树种植地块较为零碎,总的果树种植面积为6 838 hm²。在验证样本的基础上,通过混淆矩阵计算提取结果的精度,结果显示,Kappa系数为0.87,果树种植区提取的用户精度和制图精度分别为92.91%和90.77%。结果说明,本文所提出的方法适用于大区域果树的遥感提取,可为基于中高分辨率遥感影像的果树种植区监测提供有效的技术手段。     

基于Bi-LSTM模型的时间序列遥感作物分类研究

【目的】及时、准确地作物分类制图是农情监测的重要依据。本研究基于双向长短期记忆网络模型探究深度学习技术在时间序列遥感作物分类与早期识别中的应用潜力。【方法】本文以黄河三角洲地区为例，以哨兵2号全年可用卫星影像为数据源，构建年时间序列NDVI数据集;采用循环神经网络构架，搭建针对结构化时序数据的双向长短期记忆网络模型（bidirectional long short-term memory，Bi-LSTM），开展遥感作物分类，并评估模型的泛化能力;通过输入不同长度时间序列遥感数据，探究满足一定制图精度条件下的作物最早可识别时间。【结果】作物年生长时序特征对于大多数作物遥感分类识别都具有较好的区分能力，基于年时间序列NDVI数据的Bi-LSTM模型作物分类总体准确率达90.9%，Kappa系数达到0.892。通过测试不同时间序列长度对作物分类的影响发现，对大多数作物来说，其分类精度随着数据时间序列长度增加而不断提高，冬小麦、水稻等作物在生长季早期即具有较为独特的分类特征，因而利用生长季早期的时间序列影像即可获得较高的制图精度，而棉花、春玉米等作物需要完整生长序列影像才能更好地保证分类精度。【结论】卫星影像时间序列蕴含的结构化特征信息可以有效地降低特定时段的作物光谱混淆;双向循环神经网络模型能够同时考虑前向和后向的时间状态信息，可以学习作物不同阶段的光谱变化特征，在水稻、棉花、春玉米等易混淆作物的识别上表现优异;模型能够有效地把握样本总体上的变化趋势，在农作物多分类任务中表现出较好的泛化能力和鲁棒性。本研究通过集成深度学习和遥感时间序列，为及时、快速的区域作物高精度制图提供了可行的思路。     

大理苍山东西坡植被的垂直分布格局

  目的  分析云南省大理苍山东西坡植被的垂直分布格局变化特征,为有效保护苍山生态环境和物种多样性提供参考依据。  方法  以大理苍山为研究区域,基于高分二号(GF-2)高分辨率遥感影像,结合大理苍山完整的山地植被垂直地带性分布规律,辅以纹理特征和数字高程模型(DEM)数据,采用面向对象的多层次图像分割法,通过构建地形约束因子参与分类过程,准确选择样本,高精度提取研究区域的植被信息,并分析苍山东西坡植被的垂直分布格局。  结果  ①引入辅助信息的面向对象分类法提取的苍山各植被类型连续且效果好,分类总体精度为95.3%,Kappa系数为0.946 6。②苍山东西坡现状植被垂直分布格局明显,各自具有6个垂直分布带,并随着海拔高程的增大,植被分布类型趋同性增大,但东西坡垂直带谱内的优势植被类型相比也存在部分差异。  结论  相较于传统主观性强的分类方法,引入垂直带谱信息的地形约束因子进行分类,可以有效地提高山地植被分类的精度。基于面向对象的多层次分割法适用于苍山植被信息的精确提取。图5表3参20