Hyperspectral
Last updated
Last updated
The Headwall HP is a line-scan, or "pushbroom," hyperspectral sensor. Line-scan cameras collect data as a single, one-dimensional row of pixels. To build a two-dimensional image, motion is required. In conjunction with UAS technology, this motion is created through the forward flight of a UAV, creating a two-dimensional image one row of pixels at a time. In exchange for losing spatial dimensionality, in the form of only capturing a one-dimensional spatial image, line-scan hyperspectral sensors gain the advantage of adding additional spectral dimensionality.
Because the spatial dimension of push-broom hyperspectral sensors is only a single pixel tall, any horizontal movement of the sensor causes image distortion and potentially artifacting. Additionally, rotation of the sensor also introduces stretching and artifacts in the image. Finally, change in imaging speed around this axis can cause distortion in the image. Typically, these phenomena are able to be corrected for or reduced through GRYFN's orthorectification technique in the post-processing stage. However, this means that on a UAV line scan cameras are very sensitive to vibrations, wind conditions, and flight speed. See the #sensor-framerate section to learn more about operating techniques to limit the impacts of camera instability.
Hyperspectral cameras are very sensitive to lighting conditions. Illumination in the scene must remain constant, and bright. It is recommended to only fly in a ± 2-hour window of solar noon and with clear sky conditions. This is the brightest and most consistent sun conditions in the day. In the winter when sun angles are generally lower, this window should be ± 1 hour.
Frame period is the setting that determines how long each data frame should take to capture. Exposure time is the sensor setting that impacts how much of that time the sensor should be exposed to light. As exposure time increases, the sensor is exposed to light for a longer period of time, thus increasing the amount of energy observed. One important note about the relationship between these two settings is that Frame Period must always be greater than Exposure. Certain sensors will also build in an additional delay between these two settings for the frame readout time, or the time required to digitalize and write the pixel to disk.
In addition to the impact of clouds on hyperspectral data, long sun angles (observed outside of the "solar noon window") produce shadows on imagery similar to clouded conditions. This causes minuscule energy reflectance from objects of interest.
Unlike frame cameras, very little overlap between adjacent flight lines, also known as side-overlap or "sidelap" is required. Frame cameras, such as RGB or Thermal, rely on matching features in spatially adjacent images in order to stitch together a mosaic. Line-scan cameras simply overlay adjacent flight line data together, using GNSS/IMU data to determine the camera position. This process attempts to select the "best" pixels, or pixels from as close to the center of the collection as possible, to reduce any distortion that may be observed near the horizontal edges of the data cube.
The only sidelap required for line-scan cameras comes from instability in the roll of the aircraft. Generally, a minimum of just 30% sidelap can account for this aircraft roll instability. Whereas frame cameras require a minimum of 75% overlap and sidelap to guarantee enough feature similarity between images to stitch a mosaic.
Line-scan cameras collect data at a fixed framerate. Stable flight speed is greatly beneficial to creating a high-quality "data cube." As flight speed increases beyond the limit of the fixed framerate, the image scene will be under sampled as the camera cannot keep up with the greater movement of the camera without missing rows of data spatially.
Flight speed decreasing beneath the limit of the fixed framerate results in oversampled data, where a single spatial row of data is captured more than once. This can be easily accounted for in the mosaicing and orthorectification process. Generally, this means that oversampling data collection is preferred. The framerate of the camera should be set such that data is either properly sampled compared to framerate, or even oversampled.
In practice, the framerate or frame period of line-scan cameras should be set to match the flight speed or set such that the flight speed is slower than the accounted for framerate. To demonstrate numerically, at a 40m flight altitude and 4m/s flight speed, the Headwall NanoHP with a 12.6mm lens requires a 4.65ms frame period (215 Hz). Thus, a faster frame period (lower number), such as 4.43ms, should be used. This number comes from a frame period estimation using a 4.2m/s flight speed. Generally, at flight speeds at or below 5m/s, a 0.2m/s difference in estimated flight speed is sufficient to properly sample data.
