A patch‐based image classification by integrating hyperspectral data with GIS

Hyperspectral remote sensing data provide detailed spectral information and are widely used for pixel‐based image classification. However, without considering spatial correlation among neighbouring pixels, a generated thematic map may have a ‘salt‐and‐pepper’ appearance. With the development of the Geographic Information System (GIS), the spatial relationship between a pixel and its neighbours can be recorded readily and used together with remote sensing data. The objective of this study was to integrate hyperspectral data with the GIS for effective thematic mapping. To date, GIS data have been used mainly in field surveys or training field selection for remote sensing data interpretation. Here we propose a patch‐classification based on integration of the GIS with remote sensing data. The classification results obtained by using this method can be easily saved in a vector format as used for GIS files. Computational cost is decreased compared with a pixel‐by‐pixel classification. The issue of how to identify pure or mixed patches is addressed and a three‐level simple and effective checking method is developed. A case study is presented with a hyperspectral data set recorded by the Pushbroom Hyperspectral Imager (PHI) and related GIS data.     

A multiresolution spectral angle‐based hyperspectral classification method

Due to the lack of training samples, hyperspectral classification often adopts the minimum distance classification method based on spectral metrics. This paper proposes a novel multiresolution spectral‐angle‐based hyperspectral classification method, where band subsets will be selected to simultaneously minimize the average within‐class spectral angle and maximize the average between‐class spectral angle. The method adopts a pairwise classification framework (PCF), which decomposes the multiclass problem into two‐class problems. Based on class separability criteria, the original set of bands is recursively decomposed into band subsets for each two‐class problem. Each subset is composed of adjacent bands. Then, the subsets with high separability are selected to generate subangles, which will be combined to measure the similarity. Following the PCF, the outputs of all the two‐class classifiers are combined to obtain the final output. Tested with an Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) data set for a six‐class problem, the results demonstrate that our method outperforms the previous spectral metric‐based classification methods.     

Fuzzy segmentation for object‐based image classification

This Letter proposes an object‐based image classification procedure which is based on fuzzy image‐regions instead of crisp image‐objects. The approach has three stages: (a) fuzzification in which fuzzy image‐regions are developed, resulting in a set of images whose digital values express the degree of membership of each pixel to target land‐cover classes; (b) feature analysis in which contextual properties of fuzzy image‐regions are quantified; and (c) defuzzification in which fuzzy image‐regions are allocated to target land‐cover classes. The proposed procedure is implemented using automated statistical techniques that require very little user interaction. The results indicate that fuzzy segmentation‐based methods produce acceptable thematic accuracy and could represent a viable alternative to current crisp image segmentation approaches.     

Phase correlation pixel‐to‐pixel image co‐registration based on optical flow and median shift propagation

With singular value decomposition (SVD) and robust 2‐dimensional fitting phase correlation algorithms, it is possible to achieve pixel‐to‐pixel image co‐registration at sub‐pixel accuracy via local feature matching. However, the method often fails in featureless and low correlation areas making it not robust for co‐registration of images with considerable spectral differences and large featureless ground objects. A median shift propagation (MSP) technique is proposed to eliminate the problem, in a phase correlation and Normalized Cross‐Correlation (NCC) combined approach. The experiment results using images from different sensor platforms and spectral bands indicate that the new method is very robust to featureless and low correlation areas and can achieve very accurate pixel‐to‐pixel image co‐registration with good tolerance of spectral and spatial differences between images. The method will significantly improve change detection in various remote sensing applications.     

RVM‐based multi‐class classification of remotely sensed data

The relevance vector machine (RVM), a Bayesian extension of the support vector machine (SVM), has considerable potential for the analysis of remotely sensed data. Here, the RVM is introduced and used to derive a multi‐class classification of land cover with an accuracy of 91.25%, a level comparable to that achieved by a suite of popular image classifiers including the SVM. Critically, however, the output of the RVM includes an estimate of the posterior probability of class membership. This output may be used to illustrate the uncertainty of the class allocations on a per‐case basis and help to identify possible routes to further enhance classification accuracy.     

Updating land cover classification using a rule‐based decision system

This paper proposes a land cover classification methodology in agricultural contexts that provides satisfactory results with a single satellite image per year acquired during the growing period. Our approach incorporates ancillary data such as cropping history (inter‐annual crop rotations), context (altitude, soil type) and structure (parcels size and shape) to compensate for the lack of radiometric data resulting from the use of a single image. The originality of the proposed method resides in the three successive steps used: S1: per‐pixel classification of a single SPOT XS image with a restricted number of land cover classes (RL) chosen to ensure good accuracy; S2: conversion of RLs into a per‐parcel classification system using ancillary parcel boundaries; and S3: parcel allocation using exhaustive land cover classes (EL) and its refinement through the application of decision rules. The method was tested on a 120?km² area (Sousson river basin, Gers, France) where exhaustive knowledge of land cover for two successive years allowed complete validation of our method. It allocated 87% of the parcels with a 75% accuracy rate according to the exhaustive list (EL). This is a satisfactory result obtained with one SPOT XS image in a small agricultural parcel context.     

Yield prediction and waterlogging assessment for tea plantation land using satellite image‐based techniques

Motivated by the operational use of remote sensing in various agricultural crop studies, this study evaluates the application and utility of remote sensing‐based techniques in yield prediction and waterlogging assessment of tea plantation land in the Assam State of India. The potential of widely used vegetation indices like NDVI and SR (simple ratio) and the recently proposed TVI has been evaluated for the prediction of green leaf tea yield and made tea yield based on image‐derived leaf area index (LAI), along with weather parameters. It was observed that the yield model based on the TVI showed the highest correlation (R² = 0.83) with green leaf tea yield. The NDVI‐ and SR‐based models suffered non‐responsiveness when the yield approached maximum. The NDVI and SR showed saturation when the LAI exceeded a magnitude of 4. However, the TVI responded well, even when the LAI exceeded 5, and thus has potential use in the estimation of the LAI of dense vegetation such as some crops and forest where it generally exceeds the threshold value of 4.

An attempt was made for the innovative application of TCT and NDWI in the mapping of waterlogging in tea plantation land. The NDWI in conjunction with TCT offered fairly good accuracy (87%) in the delineation of tea areas prone to waterlogging. This observation indicates the potential of NDWI and TCT in mapping waterlogged areas where the soil has considerable vegetation cover.     

Examining pine spectral separability using hyperspectral data from an airborne sensor: An extension of field‐based results

Three southern USA forestry species, loblolly pine (Pinus taeda), Virginia pine (Pinus virginiana), and shortleaf pine (Pinus echinata), were previously shown to be spectrally separable (83% accuracy) using data from a full‐range spectroradiometer (400–2500 nm) acquired above tree canopies. This study focused on whether these same species are also separable using hyperspectral data acquired using the airborne visible/infrared imaging spectrometer (AVIRIS). Stepwise discriminant techniques were used to reduce data dimensionality to a maximum of 10 spectral bands, followed by discriminant techniques to measure separability. Discriminatory variables were largely located in the visible and near‐infrared regions of the spectrum. Cross‐validation accuracies ranged from 65% (1 pixel radiance data) to as high as 85% (3×3 pixel radiance data), indicating that these species have strong potential to be classified accurately using hyperspectral data from air‐ or space‐borne sensors.     

Improved topographic correction of forest image data using a 3‐D canopy reflectance model in multiple forward mode

In most forestry remote sensing applications in steep terrain, simple photometric and empirical (PE) topographic corrections are confounded as a result of stand structure and species assemblages that vary with terrain and the anisotropic reflective properties of vegetated surfaces. To address these problems, we present MFM‐TOPO as a new physically‐based modelling (PBM) approach for normalising topographically induced signal variance as a function of forest stand structure and sub‐pixel scale components. MFM‐TOPO uses the Li‐Strahler geometric optical mutual shadowing (GOMS) canopy reflectance model in Multiple Forward Mode (MFM) to account for slope and aspect influences directly. MFM‐TOPO has an explicit physical‐basis and uses sun‐canopy‐sensor (SCS) geometry that is more appropriate than strictly terrain‐based corrections in forested areas since it preserves the geotropic nature of trees (vertical growth with respect to the geoid) regardless of terrain, view and illumination angles. MFM‐TOPO is compared against our recently developed SCS+C correction and a comprehensive set of other existing PE and SCS methods (cosine, C correction, Minnaert, statistical‐empirical, SCS, and b correction) for removing topographically induced variance and for improving SPOT image classification accuracy in a Rocky Mountain forest in Kananaskis, Alberta Canada. MFM‐TOPO removed the most terrain‐based variance and provided the greatest improvement in classification accuracy within a species and stand density based class structure. For example, pine class accuracy was increased by 62% over shaded slopes, and spruce class accuracy was increased by 13% over more moderate slopes. In addition to classification, MFM‐TOPO is suitable for retrieving biophysical parameters in mountainous terrain.     

Comparison of pixel‐based and object‐oriented image classification approaches—a case study in a coal fire area,Wuda, Inner Mongolia,China

Pixel‐based and object‐oriented classifications were tested for land‐cover mapping in a coal fire area. In pixel‐based classification a supervised Maximum Likelihood Classification (MLC) algorithm was utilized; in object‐oriented classification, a region‐growing multi‐resolution segmentation and a soft nearest neighbour classifier were used. The classification data was an ASTER image and the typical area extent of most land‐cover classes was greater than the image pixels (15 m). Classification results were compared in order to evaluate the suitability of the two classification techniques. The comparison was undertaken in a statistically rigorous way to provide an objective basis for comment and interpretation. Considering consistency, the same set of ground data was used for both classification results for accuracy assessment. Using the object‐oriented classification, the overall accuracy was higher than the accuracy obtained using the pixel‐based classification by 36.77%, and the user’s and producer’s accuracy of almost all the classes were also improved. In particular, the accuracy of (potential) surface coal fire areas mapping showed a marked increase. The potential surface coal fire areas were defined as areas covered by coal piles and coal wastes (dust), which are prone to be on fire, and in this context, indicated by the two land‐cover types ‘coal’ and ‘coal dust’. Taking into account the same test sites utilized, McNemar’s test was used to evaluate the statistical significance of the difference between the two methods. The differences in accuracy expressed in terms of proportions of correctly allocated pixels were statistically significant at the 0.1% level, which means that the thematic mapping result using object‐oriented image analysis approach gave a much higher accuracy than that obtained using the pixel‐based approach..     

Object‐oriented and textural image classification of the Siberia GBFM radar mosaic combined with MERIS imagery for continental scale land cover mapping

Boreal forests and wetlands play an important role in the climate system, in particular through biosphere–atmosphere flux exchanges. They are an important pool of carbon and their role as sink or source of greenhouse gases is not fully understood. Accurate mapping of the vegetation of Siberia can therefore contribute to a better understanding of these processes at regional scale and of their effects on the climate through regional biosphere modeling. The potential of the combination of radar data with medium‐resolution optical data to obtain regional‐scale land cover mapping is investigated using multi‐spectral imagery from the MERIS sensor at 300 m resolution and a high resolution radar mosaic (pixel spacing of 100 m) covering Western and Eastern Siberia compiled in the framework of the Global Boreal Forest Mapping project. For this purpose, capabilities of oriented‐object image analysis associated to wavelet multi‐resolution techniques are investigated. Results show that wavelet multi‐resolution textures bring relevant additional information for land cover classification. Suggestions are made for the implementation of an object‐based wavelet multi‐resolution texture estimator.