Bradley A. Wilson

Pixel sampling of remotely sensed digital imagery

Abstract A full range of sampling strategies is required to support the analysis of remotely sensed data and other geographically referenced digital data in image format. Random samples generated from coordinate lists and compiled in attribute tables are critical in the selection of classifier algorithms and mapping variables, training and test area field site selection, and determination of classification accuracy. This paper introduces a number of programs which were written to implement a comprehensive sampling package in two different operating environments. Image data are extracted as samples that can be stratified by class or other variable, and can be sorted into training and test pixel tables. An illustration of the programs in classification accuracy assessment of a SPOT satellite image for classification of vegetation patterns in subarctic Canada is used to highlight the functionality of the system.

Spatial and spectral classification of remote-sensing imagery

Abstract The use of spatial satellite image information and digital elevation models in remote-sensing classification is described for a mountainous region in southwestern Yukon. A three-stage classification method was devised that incorporates a quadtree-based segmentation operator, a Gaussian minimum distance to means test, and a final test involving ancillary topographic data and a spectral curve measure. The overall improvement in accuracy is significant compared to simple multispectral techniques, and the resulting map products are consistent with few unclassified areas. The three-stage classifier can produce an output map in significantly less time than that required for per-pixel maximum likelihood classifiers, and uses a minimum of field or training data which may be difficult and expensive to acquire in complex terrain. Programs to handle spatial and spectral attributes are coded efficiently in the C programming language. They can be adapted to locate homogeneous regions in high resolution aerial imaging spectrometer data sets (down to 0.1 m pixel resolution) or other raster databases.

Topographic dependence of synthetic aperture radar imagery

Abstract The increasing availability of synthetic aperture radar (SAR) remote-sensing imagery for earth-science applications creates the need for reliable computer methods to improve the relationships between SAR observations and the Earths abiotic, biotic, and cultural resources. In this paper, the topographic effect on aerial and satellite SAR imagery is quantified and corrected using software modified from an earlier normalized-cosine package written for use with optical/infrared remote-sensing imagery. The basic idea is that the incidence angle and forest canopy interactions with the radar beam can be estimated using a digital elevation model (DEM) and near-coincident observations in the red and near-infrared portions of the spectrum. Four different study areas in Canada and three different types of SAR imagery are used to illustrate the topographic dependence, and the degree of accuracy that can be expected, following the application of these relatively simple radiometric corrections.

Characterization of Alpine Vegetation Cover Using Satellite Remote Sensing in the Front Ranges, St. Elias Mountains, Yukon Territory

Two satellite images were acquired from the SPOT High Resolution Visible (HRV) Multispectral Linear Array (MLA) sensor in 1990 and the Landsat Thematic Mapper (TM) sensor in 1985 for a mountainous region in southwest Yukon Territory, Canada. These data were classified using a maximum likelihood decision rule and a new three-stage classifier designed specifically for analysis of mountain environments. Classification results improved from approximately 74% correct when using satellite sensor data alone to 85% correct using the satellite sensor data with geomorphometric variables extracted from a digital elevation model. Further refinement to approximately 88% correct were obtained with the new classifier. The resulting image maps were used to interpret the extent of changes in vegetation cover resulting from fluctuations in river systems and groundwater regimes.

Minimizing the negative effect of the overlapping pixels on the classification accuracy of the error back-propagation neural network classifier using the ancillary and supplemental data

A procedure is presented for minimizing the problems associated with the overlapping land cover spectral signatures on the classification accuracy, which involves the use of ancillary and supplemental data in remote sensing classification. Multispectral remote sensing data from a section of Kananaskis Country in the Rocky Mountains in Canada were used as an application for this investigation. A technique was established to simplify the method of selecting the dataset of the spectral training sample that can be used to determine decision rules for classifying the application area and improving the sorting process between the overlapping pixels. A three-stage computer classifier model was developed based on the back-propagation artificial neural network method to enhance the sorting process among land cover types that have similar spectral signatures. The performance of this computer model was assessed by comparing the results with the results of three prepackaged computer classifiers using the same dataset: maximum likelihood, minimum distance to means, and parallelepiped. The investigation demonstrated that the developed computer model performed satisfactorily overall with the use of the proposed enhanced sorting procedure.

Constant time 0(1) pixel averaging with applicability to kernel filtering

The standard method of programming kernel-based image processing tasks requires an exponential increase in processing time as kernel size increases. A new approach to programming kernel-based image processing tasks is presented that requires near constant time to process a given image at any given kernel size. A comparison is made using a standard method of average filtering and a new fast algorithm. Each algorithm was tested five times and processing times were averaged for kernel sizes ranging from 3 × 3 to 101 × 101. Results show the new algorithm is faster at all kernel sizes and orders of magnitude faster at larger kernel sizes (e.g., nearly a thousand times faster when using a 101 × 101 kernel size). A discussion is provided on other types of kernel-based image processing tasks that can also use this same programming approach.

Mapping Food Desert Persistency in Thunder Bay, Ontario, 1996-2006

Thunder Bay, Ontario is a remote community that has been identified in recent government reports as having high percentage of residents living below the poverty line as well as having the largest increase in the province in food bank usage from 2007 to 2009. Identifying neighbourhoods with multiple risk factors and poor access to full-service food retail provision (i.e., food deserts) is a key step in strengthening the local food system. Food deserts were found in all three census years used in this study. Two food deserts persisted and many new food deserts appeared on peripheral regions of the city over this time period.