Siamak Khorram

The effects of image misregistration on the accuracy of remotely sensed change detection

Image misregistration has become one of the significant bottlenecks for improving the accuracy of multisource data analysis, such as data fusion and change detection. In this paper, the effects of misregistration on the accuracy of remotely sensed change detection were systematically investigated and quantitatively evaluated. This simulation research focused on two interconnected components. In the first component, the statistical properties of the multispectral difference images were evaluated using semivariograms when multitemporal images were progressively misregistered against themselves and each other to investigate the band, temporal, and spatial frequency sensitivities of change detection to image misregistration. In the second component, the ellipsoidal change detection technique, based on the Mahalanobis distance of multispectral difference images, was proposed and used to progressively detect the land cover transitions at each misregistration stage for each pair of multitemporal images. The impact of misregistration on change detection was then evaluated in terms of the accuracy of change detection using the output from the ellipsoidal change detector. The experimental results using Landsat Thematic Mapper (TM) imagery are presented. It is interesting to notice that, among the seven TM bands, band 4 (near-infrared channel) is the most sensitive to misregistration when change detection is concerned. The results from false change analysis indicate a substantial degradation in the accuracy of remotely sensed change detection due to misregistration. It is shown that a registration accuracy of less than one-fifth of a pixel is required to achieve a change detection error of less than 10%.

A feature-based image registration algorithm using improved chain-code representation combined with invariant moments

A new feature-based approach to automated image-to-image registration is presented. The characteristic of this approach is that it combines an invariant-moment shape descriptor with improved chain-code matching to establish correspondences between the potentially matched regions detected from the two images. It is robust in that it overcomes the difficulties of control-point correspondence by matching the images both in the feature space, using the principle of minimum distance classifier (based on the combined criteria), and sequentially in the image space, using the rule of root mean-square error (RMSE). In image segmentation, the performance of the Laplacian of Gaussian operators is improved by introducing a new algorithm called thin and robust zero crossing. After the detected edge points are refined and sorted, regions are defined. Region correspondences are then performed by an image-matching algorithm developed in this research. The centers of gravity are then extracted from the matched regions and are used as control points. Transformation parameters are estimated based on the final matched control-point pairs. The algorithm proposed is automated, robust, and of significant value in an operational context. Experimental results using multitemporal Landsat TM imagery are presented.

Regional Scale Land Cover Characterization Using MODIS-NDVI 250 m Multi-Temporal Imagery: A Phenology-Based Approach

Currently available land cover data sets for large geographic regions are produced on an intermittent basis and are often dated. Ideally, annually updated data would be available to support environmental status and trends assessments and ecosystem process modeling. This research examined the potential for vegetation phenology-based land cover classification over the 52,000 km2 Albemarle-Pamlico estuarine system (APES) that could be performed annually. Traditional hyperspectral image classification techniques were applied using MODIS-NDVI 250 m 16-day composite data over calendar year 2001 to support the multi-temporal image analysis approach. A reference database was developed using archival aerial photography that provided detailed mixed pixel cover-type data for 31,322 sampling sites corresponding to MODIS 250 m pixels. Accuracy estimates for the classification indicated that the overall accuracy of the classification ranged from 73% for very heterogeneous pixels to 89% when only homogeneous pixels were examined. These accuracies are comparable to similar classifications using much higher spatial resolution data, which indicates that there is significant value added to relatively coarse resolution data though the addition of multi-temporal observations.

A comparison of SPOT and Landsat-TM data for use in conducting inventories of forest resources

Abstract SPOT multispectral (XS) and Landsat Thematic Mapper (TM) digital data were studied in an attempt to evaluate the use of this data in detailed assessments of forest conditions. Forest type, basal area, and age class information were collected from 256 sample sites within an intensively managed 80000acre experimental forest in North Carolina, U.S.A. A comparison of the SPOT and TM data with the sample site information showed that XS3, the near-infrared waveband, and TM bands 2, 3, 4, 5, and 7 were significantly correlated with basal area. Age class was not found to be significantly correlated with any of the three SPOT XS wavebands. TM bands 2, 3, 4, 5, and 7 were, however, shown to be significantly correlated with age class. Although significant, the correlation coefficients between the TM or SPOT waveband data and basal area or age class were low (<0.65). Six forest cover types, and an additional water category, were selected as the basis of a land cover classification system for use with the TM ...

An Automated Artificial Neural Network System for Land Use/Land Cover Classification from Landsat TM Imagery

This paper focuses on an automated ANN classification system consisting of two modules: an unsupervised Kohonen’s Self-Organizing Mapping (SOM) neural network module, and a supervised Multilayer Perceptron (MLP) neural network module using the Backpropagation (BP) training algorithm. Two training algorithms were provided for the SOM network module: the standard SOM, and a refined SOM learning algorithm which incorporated Simulated Annealing (SA). The ability of our automated ANN system to perform Land-Use/Land-Cover (LU/LC) classifications of a Landsat Thematic Mapper (TM) image was tested using a supervised MLP network, an unsupervised SOM network, and a combination of SOM with SA network. Our case study demonstrated that the ANN classification system fulfilled the tasks of network training pattern creation, network training, and network generalization. The results from the three networks were assessed via a comparison with reference data derived from the high spatial resolution Digital Colour Infrared (CIR) Digital Orthophoto Quarter Quad (DOQQ) data. The supervised MLP network obtained the most accurate classification accuracy as compared to the two unsupervised SOM networks. Additionally, the classification performance of the refined SOM network was found to be significantly better than that of the standard SOM network essentially due to the incorporation of SA. This is mainly due to the SA-assisted classification utilizing the scheduling cooling scheme. It is concluded that our automated ANN classification system can be utilized for LU/LC applications and will be particularly useful when traditional statistical classification methods are not suitable due to a statistically abnormal distribution of the input data.

Hierarchical maximum-likelihood classification for improved accuracies

Among the supervised parametric classification methods, the maximum-likelihood (MLH) classifier has become popular and widespread in remote sensing. Reliable prior probabilities are not always freely available, and it is a common practice to perform the MLH classification with equal prior probabilities. When equal prior probabilities are used, the advantages in MLH classification may not be attained. This study has explored a hierarchical pixel classification (HPC) method to estimate prior probabilities for the spectral classes from the Landsat thematic mapper (TM) data and spectral signatures. The TM pixels were visualized in multidimensional feature space relative to the spectral class probability surfaces. The pixels that fell within more than one probability region or outside all probability regions were categorized as the pixels likely to misclassify. Prior probabilities were estimated from the pixels that fell within spectral class probability regions. The pixels most likely to be correctly classified do not need extra information and were classified according to the probability region in which they fell. The pixels likely to be misclassified need additional information and were classified by MLH classification with the estimated prior probabilities. The classified image resulting from the HPC showed increased accuracy over three classification methods. Visualization of pixels in multidimensional feature space, relative to the spectral class probability reforms, overcome the practical difficulty in estimating prior probabilities while utilizing the available information.

Per-pixel Classification of High Spatial Resolution Satellite Imagery for Urban Land-cover Mapping

Commercial high spatial resolution satellite data now provide a synoptic and consistent source of digital imagery with detail comparable to that of aerial photography. In the work described here, per-pixel classification, image fusion, and GIS-based map refinement techniques were tailored to pan-sharpened 0.61 m QuickBird imagery to develop a six-category urban land-cover map with 89.3 percent overall accuracy ( �� 0.87). The study area was a rapidly developing 71.5 km 2 part of suburban Raleigh, North Carolina, U.S.A., within the Neuse River basin. “Edge pixels” were a source of classification error as was spectral overlap between bare soil and impervious surfaces and among vegetated cover types. Shadows were not a significant source of classification error. These findings demonstrate that conventional spectral-based classification methods can be used to generate highly accurate maps of urban landscapes using high spatial resolution imagery.

Water quality mapping of Augusta Bay, Italy from Landsat-TM data

Abstract Landsat Thematic Mapper (TM) digital data were used to map the distributions and concentrations of selected water quality indicators in and around Augusta Bay, Sicily. The general approach involved near-simultaneous aquisition of TM data and water quality samples from 42 sites, laboratory analysis of samples, extraction of sample site digital numbers from the TM data, development and validation of regression models based on sample data, application of models to the entire study area, and generation of colour-coded output maps. Results were good for modelling temperature, turbidity, Secchi disk depth and chlorophyll-a, and indicate that remotely-sensed data may be applicable to monitoring water quality in this geographic area.

Comparson of Landsat MSS and TM Data for Urban Land-Use Classification

A supervised classification of digital Landsat Multispectral Scanner (MSS) data for the Raleigh, North Carolina, metropolitan area was conducted in 1982. These data were categorized into 10 land-use/land-cover types representative of the area. Digital Landsat Thematic Mapper (TM) data, for the Raleigh metropolitan area, were obtained in 1985 and analyzed for comparison to the MSS data. A stratified classification based upon principal components analysis was applied to the TM data, classifying the data into the 10 land-use/land-cover categories used in the analysis of the MSS data. Comparison of photo-interpreted land-use types and Landsat derived land-use types indicates that TM data provides significantly higher classification accuracies than can be obtained from MSS data. However, an increase in confusion between urban cover types was observed for the classified TM data over the MSS data. It is felt that the stratified classification approach was instrumental in reducing classification errors between general land-use/land-cover types such as urban areas, coniferous forests, and deciduous forests. It is not clear that the information extracted from the TM data regarding the urban environment will be of much more use to city planners than that obtained from MSS data.

Development of a feature-based approach to automated image registration for multitemporal and multisensor remotely sensed imagery

A new feature-based approach to automated multitemporal and multisensor image registration is presented. The characteristics of this technique is that it combines moment invariant shape descriptors with modified chain code correlation to establish the correspondences between potential matched regions in two images. It also overcomes the difficulties in control point correspondence in image matching caused by the problem of feature inconsistency. In image segmentation, the authors use the improved Laplacian of Gaussian (LoG) zero-crossing edge detector. Feature matching is done in both feature space and image space based on moment invariant distance and improved chain code correlation. The centers of gravity are then extracted from matched regions and used as control points. The final transformation parameters are estimated based on the final matched control points. Experimental results using multitemporal Landsat TM imagery are presented.