Use of field reflectance data for crop mapping using airborne hyperspectral image

Recent developments in hyperspectral remote sensing technologies enable acquisition of image with high spectral resolution, which is typical to the laboratory or in situ reflectance measurements. There has been an increasing interest in the utilization of in situ reference reflectance spectra for rapid and repeated mapping of various surface features. Here we examined the prospect of classifying airborne hyperspectral image using field reflectance spectra as the training data for crop mapping. Canopy level field reflectance measurements of some important agricultural crops, i.e. alfalfa, winter barley, winter rape, winter rye, and winter wheat collected during four consecutive growing seasons are used for the classification of a HyMAP image acquired for a separate location by (1) mixture tuned matched filtering (MTMF), (2) spectral feature fitting (SFF), and (3) spectral angle mapper (SAM) methods. In order to answer a general research question “what is the prospect of using independent reference reflectance spectra for image classification”, while focussing on the crop classification, the results indicate distinct aspects. On the one hand, field reflectance spectra of winter rape and alfalfa demonstrate excellent crop discrimination and spectral matching with the image across the growing seasons. On the other hand, significant spectral confusion detected among the winter barley, winter rye, and winter wheat rule out the possibility of existence of a meaningful spectral matching between field reflectance spectra and image. While supporting the current notion of “non-existence of characteristic reflectance spectral signatures for vegetation”, results indicate that there exist some crops whose spectral signatures are similar to characteristic spectral signatures with possibility of using them in image classification.     

Object-based crop classification using multi-temporal SPOT-5 imagery and textural features with a Random Forest classifier

In this study, an object-based image analysis (OBIA) approach was developed to classify field crops using multi-temporal SPOT-5 images with a random forest (RF) classifier. A wide range of features, including the spectral reflectance, vegetation indices (VIs), textural features based on the grey-level co-occurrence matrix (GLCM) and textural features based on geostatistical semivariogram (GST) were extracted for classification, and their performance was evaluated with the RF variable importance measures. Results showed that the best segmentation quality was achieved using the SPOT image acquired in September, with a scale parameter of 40. The spectral reflectance and the GST had a stronger contribution to crop classification than the VIs and GLCM textures. A subset of 60 features was selected using the RF-based feature selection (FS) method, and in this subset, the near-infrared reflectance and the image acquired in August (jointing and heading stages) were found to be the best for crop classification.     

The effectiveness of spectral similarity measures for the analysis of hyperspectral imagery

In geological imaging spectrometry (i.e., hyperspectral remote sensing), surface compositional information (e.g., mineralogy and subsequently chemistry) is obtained by statistical comparison (by means of spectral matching algorithms) of known field- or library spectra to unknown image spectra. Though these algorithms are readily used, little emphasis has been given to comparison of the performance of the various spectral matching algorithms. Four spectral measures are presented: three that calculate the angle (spectral angle measure, SAM), the vector distance (Euclidean distance measure, ED) or the vector cross-correlation (spectral correlation measure, SCM), between a known reference and unknown target spectrum and a fourth measure that measures the discrepancy of probability distributions between two pixel vectors (the spectral information divergence, SID). The performance of these spectral similarity measures is compared using synthetic hyperspectral and real (i.e., Airborne Visible Infrared Imaging Spectrometer, AVIRIS) hyperspectral data of a (artificial or real) hydrothermal alteration system characterised by the minerals alunite, kaolinite, montmorillonite and quartz. Two statistics are used to assess the performance of the spectral similarity measures: the probability of spectral discrimination (PSD) and the power of spectral discrimination (PWSD). The first relates to the ability of the selected set of spectral endmembers to map a target spectrum, whereas the second expresses the capability of a spectral measure to separate two classes relative to a reference class. Analysis of the synthetic data set (i.e., simulated alteration zones with crisp boundaries at 1–2 nm spectral resolution) shows that (1) the SID outperforms the classical empirical spectral matching techniques (SAM, SCM and ED), (2) that SCM (SID, SAM and ED do not) exploits the overall shape of the reflectance curve and hence its outcomes are (positively and negatively) affected by the spectral range selected, (3) SAM and ED give nearly similar results and (4) for the same reason as in (2), the SCM is also more sensitive (again in positive and negative sense) to the spectral noise added. Results from the study of AVIRIS data show that SAM yields more spectral confusion (i.e., class overlap) than SID and SCM. In turn, SID is more effective in mapping the four target minerals than SCM as it clearly outperforms SCM when the target mineral coincides with the mineral phase on the ground.     

Detection of stress in tomatoes induced by late blight disease in California, USA, using hyperspectral remote sensing

Large-scale farming of agricultural crops requires on-time detection of diseases for pest management. Hyperspectral remote sensing data taken from low-altitude flights usually have high spectral and spatial resolutions, which can be very useful in detecting stress in green vegetation. In this study, we used late blight in tomatoes to illustrate the capability of applying hyperspectral remote sensing to monitor crop disease in the field scale and to develop the methodologies for the purpose. A series of field experiments was conducted to collect the canopy spectral reflectance of tomato plants in a diseased tomato field in Salinas Valley of California. The disease severity varied from stage 1 (the light symptom), to stage 4 (the sever damage). The economic damage of the crop caused by the disease is around the disease stage 3. An airborne visible infrared imaging spectrometer (AVIRIS) image with 224 bands within the wavelength range of 0.4–2.5 μm was acquired during the growing season when the field data were collected. The spectral reflectance of the field samples indicated that the near infrared (NIR) region, especially 0.7–1.3 μm, was much more valuable than the visible range to detect crop disease. The difference of spectral reflectance in visible range between health plants and the infected ones at stage 3 was only 1.19%, while the difference in the NIR region was high, 10%. We developed an approach including the minimum noise fraction (MNF) transformation, multi-dimensional visualization, pure pixels endmember selection and spectral angle mapping (SAM) to process the hyperspectral image for identification of diseased tomato plants. The results of MNF transformation indicated that the first 28 eigenimages contain useful information for classification of the pixels and the rest were mainly noise-dominated due to their low eigenvalues that had few signals. Therefore, the 28 signal eigenimages were used to generate a multi-dimensional visualization space for endmember spectra selection and SAM. Classification with the SAM technique of plants’ spectra showed that the late blight diseased tomatoes at stage 3 or above could be separated from the healthy plants while the less infected plants (at stage 1 or 2) were difficult to separate from the healthy plants. The results of the image analysis were consistent with the field spectra. The mapped disease distribution at stage 3 or above from the image showed an accurate conformation of late blight occurrence in the field. This result not only confirmed the capability of hyperspectral remote sensing in detecting crop disease for precision disease management in the real world, but also demonstrated that the spectra-based classification approach is an applicable method to crop disease identification.     

Understanding the Unique Spectral Signature of Winter Rape

Driven by significant technological developments in the hyperspectral imaging, material mapping using reference spectra has received renewed interest of the remote sensing community. The applicability of reference spectral signatures in image classification depends mainly on the material type and its spectral signature behaviour. Identification and spectral characterization of materials which exhibit unique spectral behaviour is the first step in this approach. Consequently there have been active researches for the identification of surface materials which exhibit unique spectral signatures. The uniqueness of reflectance signature of winter rape relative to its co-occurring crop species was reported in this study. Reflectance spectral libraries constructed from field spectral reflectance measurements collected over five agricultural crops (alfalfa, winter barley, winter rape, winter rye, and winter wheat) during four subsequent growing seasons were classified by the linear discriminant analysis (LDA). Further, the reference field spectral database was used for the spectral feature fitting and classification of a historical HyMAP airborne hyperspectral imagery acquired at a separate site, by spectral library search. Results indicate the existence of a meaningful spectral matching between image and field spectra for winter rape and demonstrate the potential for transferring spectral library for hyperspectral image classification. The observed consistency in the discrimination of winter rape demonstrates experimentally the fundamental principle of remote sensing which suggests the theoretical existence of unique spectral signatures for materials which can be incorporated as reference spectral signatures for hyperspectral image classification.     

基于面向对象和规则的遥感影像分类研究

讨论了面向对象和规则的光学遥感影像分类方法。首先利用多尺度分割形成影像对象，建立对象的层次结构，计算对象的光谱特征、几何特征、拓扑特征等，利用对象、特征形成分类规则，并通过不同对象层间信息的传递和合并实现对影像的分类。并以北京城市土地利用分类为例，对该方法进行了验证。     

单变量特征选择的苏北地区主要农作物遥感识别

遥感识别多源特征综合和特征优选是提高遥感影像分类精度的关键技术。农作物遥感识别中,识别特征的相对单一和数量过多均会导致作物识别精度不理想。随机森林(random forests)采用分类与回归树(CART)算法来生成分类树,结合了bagging和随机选择特征变量的优点,是一种有效的分类方法。单变量特征选择(univariate feature selection)能够对每一个待分类的特征进行测试,衡量该特征和响应变量之间的关系,根据得分舍弃不好的特征,优选得到的特征用于分类。本文基于随机森林和单变量特征选择,利用多时相光谱信息、植被指数信息、纹理信息及波段差值信息,设计多组分类实验方案,对江苏省泗洪县的高分一号(GF-1)和环境一号(HJ-1A)影像进行分类研究,旨在选择最佳的分类方案对实验区主要农作物进行识别和提取。实验结果表明:(1)多源信息综合的农作物分类精度明显高于单一的原始光谱特征分类,说明不同类型特征的引入能改善分类效果;(2)基于单变量特征选择算法的优选特征分类效果最佳,总体精度97.07%,Kappa系数0.96,表明了特征优选在降低维度的同时,也保证了较高的分类精度。随机森林和单变量特征选择结合的方法可以提高遥感影像的分类精度,为农作物的识别和提取研究提供了有效的方法。     

基于CART决策树的高分二号洞庭湖区湿地分类方法

高分卫星遥感湿地分类的关键在于解决“同物异谱、异物同谱”难题。本文将当前应用前景广泛的亚米级国产高分二号（GF-2）影像和CART决策树面向对象分类算法相结合,以湖南沅江为例进行洞庭湖区典型湿地的分类提取工作,选取包括光谱信息、几何特征、地形特征和纹理特征等多维对象特征对分类器进行训练,构建了多维特征湿地分类方法。试验区总体分类精度优于传统方法,可为基于GF-2影像的洞庭湖湿地分类提供技术参考。     

面向高光谱影像分类的显著性特征提取方法

针对高光谱影像分类问题，提出了一种显著性特征提取方法。首先，利用超像素分割算法将高光谱影像3个相邻波段分割为若干个小区域。然后，基于分割得到的小区域计算反映不同区域的显著性特征。最后，沿着光谱方向采用大小为3、步长为1的滑窗法获得所有波段的显著性特征。进一步将提取的显著性特征与光谱特征进行结合，并将结合后的特征输入到支持向量机中进行分类。利用Pavia大学、Indian Pines和Salinas 3组高光谱影像数据进行分类试验。试验结果表明，与传统的空间特征提取方法和基于卷积神经网络的高光谱影像分类方法相比，提取的显著性特征能够获得更高的高光谱影像分类精度，且结合光谱特征能够进一步提高分类精度。     

Wheat crop production estimation using satellite data

This paper reports acreage, yield and production forecasting of wheat crop using remote sensing and agrometeorological data for the 1998–99 rabi season. Wheat crop identification and discrimination using Indian Remote Sensing (IRS) ID LISS III satellite data was carried out by supervised maximum likelihood classification. Three types of wheat crop viz. wheat-1 (high vigour-normal sown), wheat-2 (moderate vigour-late sown) and wheat-3 (low vigour-very late sown) have been identified and discriminated from each other. Before final classification of satellite data spectral separability between classes were evaluated. For yield prediction of wheat crop spectral vegetation indices (RVI and NDVI), agrometeorological parameters (ET_max and TD) and historical crop yield (actual yield) trend analysis based linear and multiple linear regression models were developed. The estimated wheat crop area was 75928.0 ha. for the year 1998–99, which sowed ?2.59% underestimation with land record commissioners estimates. The yield prediction through vegetation index based and vegetation index with agrometeorological indices based models were 1753 kg/ha and 1754 kg/ha, respectively and have shown relative deviation of 0.17% and 0.22%, the production estimates from above models when compared with observed production show relative deviation of ?2.4% and ?2.3% underestimations, respectively.     

高精度作物分布图制作

中国自然条件复杂 ,农业种植结构多样 ,地块小而分散 ,利用遥感影像制作作物分布图的精度很难满足农业遥感估产的需求。该文利用目前最高分辨率的商用遥感卫星 (QuickBird)影像 ,采用面向对象的影像分析方法提取耕地种植地块图 ,结合详细的地面调查制作高精度的作物分布图 ,为农业遥感估产服务。     

综合聚类和上下文特征的高光谱影像分类

常规高光谱影像逐像素分类往往没有考虑空间相关性,分类结果未体现地物的空间关联和分布特征。为了在分类中充分利用空间特征,利用聚类信息并结合隐马尔可夫随机场模型讨论了高光谱遥感影像光谱-空间分类方法。首先,在不同特征提取方法（最小噪声分离、独立成分分析和主成分分析）下,使用不同聚类方法（k-均值、迭代自组织分析算法和模糊c-均值算法）借助隐马尔可夫随机场获取优化的分割图;然后,采用4连通区域标记法对分割区域标记生成图像对象,并根据支持向量机的逐像素分类结果采用多数投票法对图像对象进行分类;最后,借助凹槽窗口邻域滤波技术改进分类结果,削弱“椒盐”现象。该方法综合了监督分类和非监督分类的优势,通过聚类引入地物空间相关性信息,通过隐马尔可夫随机场引入上下文特征,较好地弥补了单纯基于光谱信息分类的不足。     

利用面向对象的分类方法从IKONOS全色影像中提取河流和道路

传统的基于像素的统计特征的分类方法在处理高分辨率影像的分类问题上遇到了很大的困难。本研究利用面向对象的影像分析方法对IKONOS全色影像进行了河流与道路的分类,包括利用影像对象的光谱特征的初次分类和利用子目标对象的线特征的二次分类两个过程;其中初次分类由于仅依据光谱信息,不能很好地将河流或道路与其他光谱特征相似的地物区分开,而通过引入子目标的形状特征进行二次分类,就可以准确地将河流与道路提取出来。试验结果表明,面向对象的分类方法能够满足高分辨率或纹理影像的分类需要,具有很大的应用潜力。     

Efficient paddy field mapping using Landsat-8 imagery and object-based image analysis based on advanced fractel net evolution approach

This paper proposes an efficient paddy field mapping method using object-based image analysis and a bitemporal data set acquired by Landsat-8 Operational Land Imager. In the proposed approach, image segmentation is the first step and its quality has a serious impact on the accuracy of paddy field classification. In order to improve segmentation quality, a new segmentation algorithm based on a frequently used method, fractal net evolution approach, is developed, with improvement mainly in merging criteria. In order to automate the process of scale parameter determination, an unsupervised scale selection method is utilized to determine the optimal scale parameter for the proposed image segmentation approach. After segmentation, four types of object-based features including geometric, spectral, textural, and contextual information are extracted and input into the subsequent classification procedure. By using a random forest classifier, paddy fields and nonpaddy fields are separated. The proposed image segmentation method and the final classification result are both quantitatively evaluated. Our segmentation method outperformed two popular algorithms according to three supervised evaluation criteria. The classification result with overall accuracy of 91.00% and kappa statistic of 0.82 validated the effectiveness of the proposed framework. Further analysis on feature importance indicated that spectral features made the most contribution as compared to the other three types of object-based features.     

面向对象的泰安市高分辨率影像住宅建筑物提取

基于eCognition Developer平台,以泰安市QuickBird影像为数据,采用面向对象多尺度分割、最邻近和隶属度分类,充分利用高分辨率QuickBird影像具有的丰富光谱、形状、纹理和结构等地物信息,对实验区进行分类并提取住宅建筑物信息。实验表明,与传统逐像元分类法相比,面向对象分类法有效地避免了分割区域的离散破碎,地类信息的提取更加完整、精确、高效。     

Spatial and Quantitative Comparison of Topographically Derived Different Classification Algorithms Using AWiFS Data over Himalayas,India

In recent years, the significant increase in research on spatial information is observed. Classification or clustering is one of the well-known methods in spatial data analysis. Traditionally, classifiers are generally based on per-pixel approaches and are not utilizing the spatial information within pixel, called mixels which is an important source of information to image classification. There are two foremost reasons behind the existence of mixels: (a) coarse or low spatial resolution of sensor and (b) topographic effects that recorded on optical satellite imagery due to differential terrain illuminations over rugged areas such as Himalayas. In the present study, different classification algorithms have been implemented to drive the impact of topography on them. Among various available, three algorithms for the mapping of snow cover region over north Indian Himalayas (India) are compared: (a) maximum likelihood classification (MLC) as supervised classifier; (b) k-mean clustering as unsupervised classifier; and (c) linear spectral mixing model (LSMM) as soft classifier. These algorithms have been implemented on AWiFS multispectral data, and analysis was carried out. The classification accuracy is estimated by the error matrices, and LSMM achieved higher accuracy (84.5–88.5%) as compared to MLC (81–84%) and k-mean (74–81%). The results highlight that topographically derived classifiers achieved better accuracy in mapping as compared to simple classifiers. The study has many applications in snow hydrology, glaciology and climatology of mountain topography.     

Study of fractional vegetation cover using high spectral resolution data

A field experiment was conducted to study the effect of vegetation cover on soil spectra and relationship of spectral indices with vegetation cover. Multi-date spectral measurements were carried out on twelve wheat fields. Five sets of measurements were taken during the growth period of wheat crop. Field reflectance data were collected in the range 350 to 1800 nm using ASD spectroradiometer. Analysis of data was done to select narrow spectral bands for estimation of ground cover. The ratio of reflectance from vegetation covered soil and reflectance from bare soil indicated that spectral reflectance at 670 and 710 nm are the most sensitive bands. Two bands in visible (670 and 560 nm), three bands in near infrared (710, 870 and 1100 nm) and three bands in middle infrared (1480, 1700 and 1800 nm) were found highly correlated with fractional cover. Vegetation indices developed using narrow band spectral data have been found to be better than those developed using broad- band data for estimation of ground cover.     

Remote Sensing Study of Sittampundi Anorthosite Complex,India

In the present study, The Landsat 7 ETM satellite data was collected for the Sittampundi anorthosites complex and digital image analysis was carried out. The anorthositic rocks available at Sittampundi complex is considered as an equivalent of lunar highland rocks. Hence, a remote sensing study comprises of image analysis and spectral profile analysis was carried out. The satellite data was digitally processed and generated various outputs like band combinations, color composites, stretched outputs, and PCA. The suitable processed outputs were identified for delineating the anorthosite complex. The diagnostic absorption features of reflectance spectra are the sensitive indicators of mineralogy and chemical composition of rocks, which are interest to the planetary scientists. The spectral profile of Landsat ETM plotted for pure and mixed anorthosite pixels and compared with the field and lab reflectance spectra. The percentages of image spectra vary from 30% to 60% for Sittampundi anorthosite. The spectral bands 2, 4 and 6 have low reflectance and bands 3 and 5 have high reflectance. The spectral range of bands 2,3,4,5 and 6 are 525 nm–605 nm, 630 nm–690 nm, 750 nm–900 nm, 1550 nm–1750 nm and 10400 nm–12500 nm respectively. The field spectral curve has weak absorptions at 650 nm and 1000 nm due to the iron transition absorption and low ca- pyroxene respectively available in the anorthosite, matching with the image spectra. However, hyperspectal image with narrow bandwidth could be more useful in selecting the suitable spectrum for remotely mapping the anorthosite region, as equivalent test site for lunar highland region.     

Hyperspectral image classification by a variable interval spectral average and spectral curve matching combined algorithm

Classification of hyperspectral images has been receiving considerable attention with many new applications reported from commercial and military sectors. Hyperspectral images are composed of a large number of spectral channels, and have the potential to deliver a great deal of information about a remotely sensed scene. However, in addition to high dimensionality, hyperspectral image classification is compounded with a coarse ground pixel size of the sensor for want of adequate sensor signal to noise ratio within a fine spectral passband. This makes multiple ground features jointly occupying a single pixel. Spectral mixture analysis typically begins with pixel classification with spectral matching techniques, followed by the use of spectral unmixing algorithms for estimating endmembers abundance values in the pixel. The spectral matching techniques are analogous to supervised pattern recognition approaches, and try to estimate some similarity between spectral signatures of the pixel and reference target. In this paper, we propose a spectral matching approach by combining two schemes—variable interval spectral average (VISA) method and spectral curve matching (SCM) method. The VISA method helps to detect transient spectral features at different scales of spectral windows, while the SCM method finds a match between these features of the pixel and one of library spectra by least square fitting. Here we also compare the performance of the combined algorithm with other spectral matching techniques using a simulated and the AVIRIS hyperspectral data sets. Our results indicate that the proposed combination technique exhibits a stronger performance over the other methods in the classification of both the pure and mixed class pixels simultaneously.     

一种多特征转换的高光谱影像自适应分类方法

:光谱相似性测度用来衡量像元光谱的相似程度,是高光谱影像光谱匹配分类的重要工具之一,一般通过设置阈值判断像元光谱和参考光谱是否相似来进行分类。在此基础上,本文提出了一种多特征转换的高光谱影像自适应分类方法,实现了各种光谱相似性特征和分类器相结合的一种自适应分类。实验结果表明,本文提出的方法相比于传统的SVM方法,分类的总体精度更高,还可以避免部分传统光谱匹配分类方法中需要专家经验确定分类阈值的复杂过程。