Name
Abstract | ||
|---|---|---|
Data from multispectral and hyperspectral imaging systems have been used in many applications including land cover classification, surface characterization, material identification, and spatially unresolved object detection. While these optical spectral imaging systems have provided useful data, their design and utility could be further enhanced by better understanding the sensitivities and relative roles of various system attributes; in particular, when application data product accuracy is used as a metric. To study system parameters in the context of land cover classification, an end-to-end remote sensing system modeling approach was previously developed. In this paper, we extend this model to subpixel object detection applications by including a linear mixing model for an unresolved object in a background and using object detection algorithms and probability of de- tection ( ) versus false alarm ( ) curves to characterize performance. Validations with results obtained from airborne hy- perspectral data show good agreement between model predictions and the measured data. Examples are presented which show the utility of the modeling approach in understanding the relative importance of various system parameters and the sensitivity of versus curves to changes in the system for a subpixel road detection scenario. The tremendous data volume associated with these sensors also motivates targeted, efficient instrument design and collec- tion methods to acquire the necessary data without drowning in that which is irrelevant. While a strength of modern hyperspec- tral sensors is that they collect comprehensive data that can be used in a variety of applications, for any given application much of the data may go unused. Also, data collected under some il- lumination or environmental conditions may not even contain the desired information. There is a cost to collecting these data in terms of instrument complexity, data storage, communica- tions, processing requirements, and missed opportunities to col- lect data elsewhere. It is, therefore, becoming increasingly im- portant to design and operate these sensors in ways that optimize the probability of collecting sufficient (quantity and quality), but not excessive, data to extract the desired information about a re- mote scene. Also, because the complexity of these data require automated algorithms, as opposed to analyst visual interpreta- |
Year | DOI | Venue |
|---|---|---|
2002 | 10.1109/TGRS.2002.1010896 | Geoscience and Remote Sensing, IEEE Transactions |
Keywords | Field | DocType |
geophysical signal processing,geophysical techniques,image processing,multidimensional signal processing,remote sensing,terrain mapping,analytical model,false alarm,geophysical measurement technique,hyperspectral imaging,image classification,land cover,land surface,multispectral imaging,probability of detection,spectral imaging system,subpixel object detection,surface characterization,optical imaging,indexing terms,multispectral images,system modeling,spectral imaging,linear mixed model | Object detection,Computer vision,False alarm,Spectral imaging,Remote sensing,Multispectral image,Image processing,Hyperspectral imaging,Artificial intelligence,Multispectral pattern recognition,Subpixel rendering,Mathematics | Journal |
Volume | Issue | ISSN |
40 | 5 | 0196-2892 |
Citations | PageRank | References |
35 | 5.22 | 6 |
Authors | ||
2 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| John P. Kerekes | 1 | 194 | 35.38 |
| Jerrold E. Baum | 2 | 35 | 5.56 |