Class Spectral Signatures Research Articles

Consistent Classification of Landsat Time Series with an Improved Automatic Adaptive Signature Generalization Algorithm

Classifying land cover is perhaps the most common application of remote sensing, yet classification at frequent temporal intervals remains a challenging task due to radiometric differences among scenes, time and budget constraints, and semantic differences among class definitions from different dates. The automatic adaptive signature generalization (AASG) algorithm overcomes many of these limitations by locating stable sites between two images and using them to adapt class spectral signatures from a high-quality reference classification to a new image, which mitigates the impacts of radiometric and phenological differences between images and ensures that class definitions remain consistent between the two classifications. We refined AASG to adapt stable site identification parameters to each individual land cover class, while also incorporating improved input data and a random forest classifier. In the Research Triangle region of North Carolina, our new version of AASG demonstrated an improved ability to update existing land cover classifications compared to the initial version of AASG, particularly for low intensity developed, mixed forest, and woody wetland classes. Topographic indices were particularly important for distinguishing woody wetlands from other forest types, while multi-seasonal imagery contributed to improved classification of water, developed, forest, and hay/pasture classes. These results demonstrate both the flexibility of the AASG algorithm and the potential for using it to produce high-quality land cover classifications that can utilize the entire temporal range of the Landsat archive in an automated fashion while maintaining consistent class definitions through time.

The effects of haze on the spectral and statistical properties of land cover classification

Haze occurs almost every year in Malaysia and is caused by smoke which originates from forest fire in Indonesia. It causes visibility to drop, therefore affecting the data acquired for this area using optical sensor such as that on board Landsat satellite. The effects of haze on the data can be observed from the spectral and statistical properties of land cover classification. The work presented in this thesis is meant to analyse the statistical properties of land cover classification of hazy dataset. Maximum Likelihood (ML) was found to be a preferable classification scheme in which the effects of haze can be investigated. The study made use of hazy dataset that were simulated based on real haze spectral and statistical properties. By investigating these dataset, the spectral and statistical properties of the land classes can be systematically analysed, in which showing that haze modifies the class spectral signatures and statistical properties, consequently causing the data quality to decline.

Consistent classification of image time series with automatic adaptive signature generalization

Long-term data archives such as Landsat offer the potential for understanding land cover dynamics over large areas, but limited progress has been made towards realizing this potential due to data availability and computational limitations. Those limitations are less relevant now, and there is renewed interest in developing reliable methods of automatically and consistently classifying time series of remotely sensed images. Our objective was to develop a method of automatically classifying temporally irregular time series (i.e., non-anniversary date images in consecutive years) of images with a minimum of parameterization and a priori information. In contrast to traditional signature extension methods, the automatic adaptive signature generalization procedure (AASG) adapts class spectral signatures to individual images and therefore requires no image correction procedure. Class signatures are derived from pixels with stable land cover through time. We tested the performance of AASG relative to traditional signature extension with various image corrections, and explored the sensitivity of AASG to a thresholding parameter (c) controlling stable site identification. AASG performed as well as signature extension with atmospheric correction (κ=0.68), and better than signature extension with relative (κ=0.65) and TOA reflectance (κ=0.56) image corrections for a summer–summer image pair. Additionally, we demonstrated the unique ability of AASG to adapt class signatures to phenological differences by classifying a summer–winter image pair with a modest reduction in overall accuracy (κ=0.66). Observed sensitivity to c supported the hypothesis of an optimum value yielding enough training sites to describe class spectral variability, but conservative enough to minimize contamination of signatures due to classification errors. AASG offers significant advantages over traditional signature extension, particularly for temporally irregular time series. Although we demonstrated a simple implementation, the AASG approach is flexible and we outline several refinements which stand to improve performance. This development represents significant progress towards realizing the potential of long-term data archives to gain long-term understandings of global land cover dynamics.

Hyperspectral band clustering and band selection for urban land cover classification

The aim of this study is to combine band clustering with band selection for dimensionality reduction of hyperspectral imagery. The performance of dimensionality reduction is evaluated through urban land cover classification accuracy with the dimensionality-reduced data. Different from unsupervised clustering using all the pixels or supervised clustering requiring labelled pixels, the discussed semi-supervised band clustering needs class spectral signatures only; band selection result is used as initial condition for band clustering; after clustering, a cluster selection step is applied to select clusters to be used in the following data analysis. In this article, we propose to conduct band selection by removing outlier bands in each cluster before finalizing cluster centres. The experimental results in urban land cover classification show that the proposed algorithm can further enhance support vector machine (SVM)-based classification accuracy.

Semi-supervised dimensionality reduction using orthogonal projection divergence-based clustering for hyperspectral imagery

Band clustering and selection are applied to dimensionality reduction of hyperspectral imagery. The proposed method is based on a hierarchical clustering structure, which aims to group bands using an information or similarity measure. Specifically, the distance based on orthogonal projection divergence is used as a criterion for clustering. After clustering, a band selection step is applied to select representative band to be used in the following data analysis. Different from unsupervised clustering using all the pixels or supervised clustering requiring labeled pixels, the proposed semi-supervised band clustering and selection needs class spectral signatures only. The experimental results show that the proposed algorithm can significantly outperform other existing methods with regard to pixel-based classification task.

Semisupervised Band Clustering for Dimensionality Reduction of Hyperspectral Imagery

Band clustering is applied to dimensionality reduction of hyperspectral imagery. Different from unsupervised clustering using all the pixels or supervised clustering requiring labeled pixels, the proposed semisupervised band clustering needs class spectral signatures only. After clustering, a cluster selection step is applied to select clusters to be used in the following data analysis. Initial conditions and distance metrics are also investigated to improve the clustering performance. The experimental results show that the proposed algorithm can outperform other existing methods with lower computational cost.

Unsupervised real-time constrained linear discriminant analysis to hyperspectral image classification

We have proposed a constrained linear discriminant analysis (CLDA) approach for classifying the remotely sensed hyperspectral images. Its basic idea is to design an optimal linear transformation operator which can maximize the ratio of inter-class to intra-class distance while satisfying the constraint that the different class centers after transformation are aligned along different directions. Its major advantage over the traditional Fisher's linear discriminant analysis is that the classification can be achieved simultaneously with the transformation. The CLDA is a supervised approach, i.e., the class spectral signatures need to be known a priori. But, in practice, these informations may be difficult or even impossible to obtain. So in this paper we will extend the CLDA algorithm into an unsupervised version, where the class spectral signatures are to be directly generated from an unknown image scene. Computer simulation is used to evaluate how well the algorithm performs in terms of finding the pure signatures. We will also discuss how to implement the unsupervised CLDA algorithm in real-time for resolving the critical situations when the immediate data analysis results are required.

Use of synthetic bands derived from mixing models in the multispectral classification of remote sensing images

An analysis of the use of features derived from class proportions in a pixel for the multispectral classification of reforested areas in Landsat images is performed. Through a linear mixing model, synthetic bands derived from those proportions are obtained either using the constrained or the weighted least squares procedures. The method indicates that the synthetic bands offer an alternative to well-known dimensionality reduction techniques such as principal components or canonical analysis. Furthermore, those bands provide a useful tool for visual interpretation, since they contain information that is related to physical concepts (proportions) more easily assimilated than class spectral signatures.