ARS | Publication request: Automated Registration of Hyperspectral Images for Precision Agriculture

Submitted to: Computers and Electronics in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: November 8, 2004
Publication Date: January 8, 2005
Citation: Erives, H., Fitzgerald, G.J. 2005. Automated registration of hyperspectral images for precision agriculture. Computers and Electronics in Agriculture. 47:103-119.

Interpretive Summary: Hyperspectral remote sensing allows the acquisition of dozens to hundreds of images of a single scene at discrete wavelengths of light in the visible and near-infrared portions of the spectrum. With a full-frame camera system, the acquisition of these images can take up to several seconds and when mounted on a moving aerial platform, such as a helicopter, the images will be offset when compared to each other. This makes analysis nearly impossible. To align, or co-register these images, a method called Phase Correlation is presented which allows the images to be aligned. This method accounts for differences in illumination between successive images, noise, blurring due to vibration and aircraft movement, and image vignetting or drop-off in illumination across the image caused by non-uniformity of the lens and filters. This method was tested on images acquired over USDA-ARS cotton research fields near Phoenix, AZ. Results showed that the method was very robust and successful at co-registering the images to within plus or minus one pixel across the entire set of images. Software methods to correct raw imagery are critical for hyperspectral remote sensing to be useful to scientists or managers of agricultural or natural resources.

Technical Abstract: Hyperspectral images of the Earth's surface are increasingly being acquired from aerial platforms. The dozens or hundreds of bands acquired by a typical hyperspectral sensor are acquired either through a scanning process or by collecting a sequence of images at varying wavelengths. This latter method has the advantage of acquiring coherent images of a scene at each wavelength but since it takes time to collect these images, some form of co-registration is required to build coherent image cubes. This paper discusses the application of the Phase Correlation (PC) Method to recover scaling, rotation, and translation from a hyperspectral imaging system. This approach is well suited for remotely sensed images acquired from a platform that is moving (along and across track). We were able to register images to within ± 1 pixels across entire image cubes obtained from a hyperspectral imaging system suitable for precision farming tasks.