Limits...
Mosaicing of hyperspectral images: the application of a spectrograph imaging device.

Moroni M, Dacquino C, Cenedese A - Sensors (Basel) (2012)

Bottom Line: The resulting mosaic was successively georeferenced within the WGS-84 geographic coordinate system.This paper also addresses how this information can be transferred to a push broom type spectral imaging device to build the hyperspectral cube of the area prior to land classification.Mapping allows for the identification of objects within the image and agrees well with ground-truth measurements.

View Article: PubMed Central - PubMed

Affiliation: DICEA-Sapienza University of Rome, via Eudossiana 18, Rome 00184, Italy. monica.moroni@uniroma1.it

ABSTRACT
Hyperspectral monitoring of large areas (more than 10 km(2)) can be achieved via the use of a system employing spectrometers and CMOS cameras. A robust and efficient algorithm for automatically combining multiple, overlapping images of a scene to form a single composition (i.e., for the estimation of the point-to-point mapping between views), which uses only the information contained within the images themselves is described here. The algorithm, together with the 2D fast Fourier transform, provides an estimate of the displacement between pairs of images by accounting for rotations and changes of scale. The resulting mosaic was successively georeferenced within the WGS-84 geographic coordinate system. This paper also addresses how this information can be transferred to a push broom type spectral imaging device to build the hyperspectral cube of the area prior to land classification. The performances of the algorithm were evaluated using sample images and image sequences acquired during a proximal sensing field campaign conducted in San Teodoro (Olbia-Tempio-Sardinia). The hyperspectral cube closely corresponds to the mosaic. Mapping allows for the identification of objects within the image and agrees well with ground-truth measurements.

No MeSH data available.


Related in: MedlinePlus

Map of the investigated area and spectral signatures.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC3472825&req=5

f17-sensors-12-10228: Map of the investigated area and spectral signatures.

Mentions: To map the georeferenced area, the Maximum Likelihood classifier with four classes was employed. The classifier uses the distribution of data within each region of interest to calculate n-D probability functions for each class (where n represents the number of bands being used in the classification). Each pixel is assigned to the class for which the highest probability is calculated. It is common to have pixels unclassified with this method (black areas in the classification map). The result of this mapping and the spectral libraries for each class identified by the classification procedure are presented in Figure 17. The signatures were grouped on the same graph. The signature of the building roofs is flat and equal to one since it was employed for the radiometric calibration of the hyperspectral cube. The spectral libraries for vegetation show that the classifier is able to distinguish the lawns from shrubs. Those spectral signatures present the typical features of vegetation, i.e., green peak, chlorophyll wells, red edge and NIR plateau. Their analysis allows the extraction of the indices that characterize the state of the vegetation. In conclusion, this classification is an excellent method for identifying the distribution of different types of framed ground surfaces.


Mosaicing of hyperspectral images: the application of a spectrograph imaging device.

Moroni M, Dacquino C, Cenedese A - Sensors (Basel) (2012)

Map of the investigated area and spectral signatures.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC3472825&req=5

f17-sensors-12-10228: Map of the investigated area and spectral signatures.
Mentions: To map the georeferenced area, the Maximum Likelihood classifier with four classes was employed. The classifier uses the distribution of data within each region of interest to calculate n-D probability functions for each class (where n represents the number of bands being used in the classification). Each pixel is assigned to the class for which the highest probability is calculated. It is common to have pixels unclassified with this method (black areas in the classification map). The result of this mapping and the spectral libraries for each class identified by the classification procedure are presented in Figure 17. The signatures were grouped on the same graph. The signature of the building roofs is flat and equal to one since it was employed for the radiometric calibration of the hyperspectral cube. The spectral libraries for vegetation show that the classifier is able to distinguish the lawns from shrubs. Those spectral signatures present the typical features of vegetation, i.e., green peak, chlorophyll wells, red edge and NIR plateau. Their analysis allows the extraction of the indices that characterize the state of the vegetation. In conclusion, this classification is an excellent method for identifying the distribution of different types of framed ground surfaces.

Bottom Line: The resulting mosaic was successively georeferenced within the WGS-84 geographic coordinate system.This paper also addresses how this information can be transferred to a push broom type spectral imaging device to build the hyperspectral cube of the area prior to land classification.Mapping allows for the identification of objects within the image and agrees well with ground-truth measurements.

View Article: PubMed Central - PubMed

Affiliation: DICEA-Sapienza University of Rome, via Eudossiana 18, Rome 00184, Italy. monica.moroni@uniroma1.it

ABSTRACT
Hyperspectral monitoring of large areas (more than 10 km(2)) can be achieved via the use of a system employing spectrometers and CMOS cameras. A robust and efficient algorithm for automatically combining multiple, overlapping images of a scene to form a single composition (i.e., for the estimation of the point-to-point mapping between views), which uses only the information contained within the images themselves is described here. The algorithm, together with the 2D fast Fourier transform, provides an estimate of the displacement between pairs of images by accounting for rotations and changes of scale. The resulting mosaic was successively georeferenced within the WGS-84 geographic coordinate system. This paper also addresses how this information can be transferred to a push broom type spectral imaging device to build the hyperspectral cube of the area prior to land classification. The performances of the algorithm were evaluated using sample images and image sequences acquired during a proximal sensing field campaign conducted in San Teodoro (Olbia-Tempio-Sardinia). The hyperspectral cube closely corresponds to the mosaic. Mapping allows for the identification of objects within the image and agrees well with ground-truth measurements.

No MeSH data available.


Related in: MedlinePlus