Peijun Li

Multispectral image segmentation by a multichannel watershed-based approach

Watershed transformation in mathematical morphology is a powerful morphological tool for image segmentation that is usually defined for greyscale images and applied to the gradient magnitude of an image. This paper presents an extension of the watershed algorithm for multispectral image segmentation. A vector‐based morphological approach is proposed to compute gradient magnitude from multispectral imagery, which is then input into watershed transformation for image segmentation. The gradient magnitude is obtained at multiple scales. After an automatic elimination of local irrelevant minima, a watershed transformation is applied to segment the image. The segmentation results were evaluated and compared with other multispectral image segmentation methods, in terms of visual inspection, and object‐based image classification using high resolution multispectral images. The experimental results indicate that the proposed method can produce accurate segmentation results and higher classification accuracy, if the scales and contrast parameter are appropriately selected in the gradient computation and subsequent local minima elimination. The proposed method shows encouraging results and can be used for segmentation of high resolution multispectral imagery and object based classification.

Urban building damage detection from very high resolution imagery using OCSVM and spatial features

The availability of commercial very high resolution (VHR) satellite imagery makes it possible to detect and assess building damage in the aftermath of earthquake disasters using these data. Although conventional change detection methods may be used to assess the building damage, the analysis is directed to all classes, both damaged and undamaged, but is not focused on the class of interest. In this paper, we proposed to detect the building damage in urban environments from multitemporal VHR image data using the One-Class Support Vector Machine (OCSVM), a recently developed one-class classifier, which requires training samples from the building damage only. This was illustrated with building damage detection in an urban environment from multitemporal Quickbird images. The detection was conducted at pixel level and object level. Different input vectors for the OCSVM classifier were tested in order to assess the discrimination power of spectral and spatial features: pixel level spectral features and texture features, as well as object‐based features. The results showed that the OCSVM performed better on the object level, with an overall accuracy of 82.33% and a kappa coefficient of 60.09%. The results also showed that the OCSVM provides a useful framework that can combine different features and focus on the building damage class of interest. More spatial features are needed to be exploited to obtain more accurate detection results in future studies.

A Multilevel Hierarchical Image Segmentation Method for Urban Impervious Surface Mapping Using Very High Resolution Imagery

This paper presents a hierarchical image segmentation method that combines multichannel watershed transformation and dynamics of watershed contours for the segmentation of very high resolution (VHR) multispectral imagery. The image gradient was first extracted from a multispectral image using a multichannel morphological method, followed by classical watershed transformation to produce an initial segmentation result. The resulting watershed contours were then analyzed according to their relevance relative to the minima of the adjacent basins to construct an image containing information about their dynamics. By thresholding the image of the contour dynamics at different levels, multilevel hierarchical segmentation results with different levels of detail were achieved. The proposed method was evaluated by comparing with existing methods through visual inspection, quantitative measures and applications in urban impervious surface mapping, using two sets of VHR image data. The experimental results showed that the proposed method produced more accurate segmentation results compared to an existing single-level segmentation method, in terms of visual and quantitative evaluations. While used for urban impervious surface mapping, the proposed method achieved an overall accuracy significantly higher than the pixel based classification method, and also higher than the object based classification using a single-level segmentation result. Compared with the most widely used segmentation method implemented in the eCognition, the proposed method achieved a comparable performance, although they have different segmentation details. The proposed segmentation method can be used in relevant VHR image processing and applications.

A super-resolution mapping method using local indicator variograms

Super-resolution mapping (SRM) is a recently developed research task in the field of remotely sensed information processing. It provides the ability to obtain land-cover maps at a finer scale using relatively low-resolution images. Existing algorithms based on indicator geostatistics and downscaling cokriging offer an SRM approach using spatial structure models derived from real data. In this article, a novel SRM method is developed based on a sequentially produced with local indicator variogram (SLIV) SRM model. In the SLIV method, indicator variograms extracted from target-resolution classification are produced from a representative local area as opposed to using the entire image. This simplifies the application of the method since limited target-resolution reference data are required. Our investigation on three diverse case studies shows that the local window (approximately 2% of the entire study area) selection process offers comparable accuracy results to those using globally derived spatial structures, indicating our methodology to be a promising practice. Furthermore, comparison of the proposed method with random realizations indicates an improvement of 7–12% in terms of overall accuracy and 15–18% in terms of the kappa coefficient. The evaluation focused on a 270–30 m pixel size reconstruction as a potential popular application, for example moving from Moderate Resolution Imaging Spectroradiometer (MODIS) to Landsat-type resolutions.

Multivariate image texture by multivariate variogram for multispectral image classification.

Traditional image texture measure usually allows a texture description of a single band of the spectrum, characterizing the spatial variability of gray-level values within the singleband image. A problem with the approach while applied to multispectral images is that it only uses the texture information from selected bands. In this paper, we propose a new multivariate texture measure based on the multivariate variogram. The multivariate texture is computed from all bands of a multispectral image, which characterizes the multivariate spatial autocorrelation among those bands. In order to evaluate the performance of the proposed texture measure, the derived multivariate texture image is combined with the spectral data in image classification. The result is compared to classifications using spectral data alone and plus traditional texture images. A machine learning classifier based on Support Vector Machines (SVMs) is used for image classification. The experimental results demonstrate that the inclusion of multivariate texture information in multispectral image classification significantly improves the overall accuracy, with 5 to 13.5 percent of improvement, compared to the classification with spectral information alone. The results also show that when incorporated in image classification as an additional band, the multivariate texture results in high overall accuracy, which is comparable with or higher than the best results from the existing single-band and two-band texture measures, such as the variogram, cross variogram and Gray-Level Co-occurrence Matrix (GLCM) based texture. Overall, the multivariate texture provides the useful spatial information for land-cover classification, which is different from the traditional single band texture. Moreover, it avoids the band selection procedure which is prerequisite to traditional texture computation and would help to achieve high accuracy in the most classification tasks.

Lithological mapping from hyperspectral data by improved use of spectral angle mapper

Abstract The spectral angle mapper (SAM), as a spectral matching method, has been widely used in lithological type identification and mapping using hyperspectral data. The SAM quantifies the spectral similarity between an image pixel spectrum and a reference spectrum with known components. In most existing studies a mean reflectance spectrum has been used as the reference spectrum for a specific lithological class. However, this conventional use of SAM does not take into account the spectral variability, which is an inherent property of many rocks and is further magnified in remote sensing data acquisition process. In this study, two methods of determining reference spectra used in SAM are proposed for the improved lithological mapping. In first method the mean of spectral derivatives was combined with the mean of original spectra, i.e., the mean spectrum and the mean spectral derivative were jointly used in SAM classification, to improve the class separability. The second method is the use of multiple reference spectra in SAM to accommodate the spectral variability. The proposed methods were evaluated in lithological mapping using EO-1 Hyperion hyperspectral data of two arid areas. The spectral variability and separability of the rock types under investigation were also examined and compared using spectral data alone and using both spectral data and first derivatives. The experimental results indicated that spectral variability significantly affected the identification of lithological classes with the conventional SAM method using a mean reference spectrum. The proposed methods achieved significant improvement in the accuracy of lithological mapping, outperforming the conventional use of SAM with a mean spectrum as the reference spectrum, and the matching filtering, a widely used spectral mapping method.

Land-Cover Change Detection using One-Class Support Vector Machine

Change detection using remote sensing has considerable potential for monitoring land-cover change. Commonly, one specific class of change is of interest in many applications. In this paper, a recently developed one-class classifier, the One-Class Support Vector Machine (OCSVM), is proposed for the change detection of one specific class by multitemporal classification. The classifier only requires samples from the change class of interest as the training data. The performance of the proposed method was evaluated in two applications by comparing with conventional post-classification comparison methods. The results demonstrated the proposed method achieved both higher overall accuracy and higher kappa coefficient than the conventional methods, and demonstrated good potential for further application. The study also indicated that with the OCVSM, the analysis can focus only on the specific class of interest and does not need to treat other classes, thus providing highly accurate change detection. The OCSVM-based change detection method, as a general and easily implemented method, can be used for applications where only the change of one specific class is of interest.

One-Class Classification of Remote Sensing Images Using Kernel Sparse Representation

Sparse representations have been widely studied in remote sensing image analysis in recent years. In this paper, we develop a novel method for one-class classification (OCC) using a kernel sparse representation model for remotely sensed imagery. Training samples taken from the target class alone are used to build a learning dictionary for the sparse representation model, which is then optimized to produce a reconstruction residual. In the proposed model, a pixel is classified as the target class if the obtained reconstruction residual for the pixel is smaller than a given threshold; otherwise, the pixel is labeled as the outlier class. To improve the data separability between the target and outliner classes, the training samples taken from the target class are mapped into a high-dimensional feature space using a kernel function to build a learning dictionary for the kernel sparse representation model. OCC is then conducted in the mapped high-dimensional feature space using the reconstruction residual threshold, following the same principle as OCC in the original feature space. The proposed OCC method is evaluated and compared with several existing OCC methods in three different case studies. The experimental results indicate that the proposed method outperforms these existing methods, particularly when using a kernel sparse representation.

Land cover classification using CHRIS/PROBA images and multi-temporal texture

Most existing multi-temporal classification studies use spectral information alone and ignore the temporal correlation between two-date images. This article proposes a new method to characterize the local temporal correlation using multi-temporal texture measured with a geostatistical function called the pseudo cross variogram (PCV). The derived multi-temporal texture, as an additional band, was combined with the spectral information in multi-temporal classification. The performance of the multi-temporal texture was evaluated and compared with the use of multi-temporal spectral data alone and plus the traditional variogram texture in land cover classification using bitemporal hyperspectral Compact High Resolution Imaging Spectrometer/Project for On Board Autonomy (CHRIS/PROBA) images. The results show that although land cover classification using spectral information from bitemporal CHRIS/PROBA data alone had an acceptable overall accuracy of 85.66%, the inclusion of multi-temporal texture in land cover classification led to significant increases (at the 95% confidence level) in both overall accuracy (3.3–4.3% improvement) and the kappa coefficient (4.9–6.6% improvement), particularly for vegetation classes. The incorporation of multi-temporal texture into multi-temporal land cover classification also outperformed the incorporation of the traditional variogram texture. The proposed method provides a new way to exploit the temporal correlation between bitemporal images for improved land cover classification.

A Novel Method for Urban Road Damage Detection Using Very High Resolution Satellite Imagery and Road Map

In this paper, a novel method for detecting earthquake damage to urban road networks using very high resolution (VHR) satellite images is proposed. The study focuses on road blockage caused by rubble from collapsed buildings. The road regions were first extracted from a pre-event road map. The damage to the road network was then detected from the post-event VHR image of the road regions using object-based classification, where both spectral and texture information was used. The damage to adjacent buildings was also extracted and used to refine the initial result of the road damage detection. The proposed method was evaluated using bitemporal VHR images and the road map of Port au Prince, Haiti, which was heavily hit by an earthquake in January 2010. The experimental results showed that the proposed method outperformed the direct use of various data combinations in road damage detection and achieved the best performance, with an overall accuracy (OA) of 81.50 percent and a Kappa coefficient of 61.23 percent. In particular, when the refinement step with building damage information in the area adjacent to the road was applied, the OA and Kappa were significantly improved.