...Study published last week in the Journal of Photogrammetry and Remote Sensing, that describes an algorithm that can classify land cover types with minimal nudging from humans.
The problem, from a computational standpoint, is that hyperspectral sensors are too good at their jobs. Where most visual data assigns a single value (like color) to each pixel, hyperspectral data pixels each have hundreds, even thousands of values (see image to the left). Statistically, this makes each pixel seem unique to the computers tasked with classification. This is known as the Hughes effect, and it’s a huge problem because it cripples the potential of using hyperspectral data to rapidly update our knowledge about the condition of the earth’s surface.
Even if they can’t label the land cover types, hyperspectral imaging algorithms are usually able to put like pixels into groups based mostly on their proximity to one another. In the new study, the authors combined this clustering method with another technique that uses a small number of training samples to label each group of pixels.
more:
http://www.sciencedirect.com/science/article/pii/S0924271614002020
Comments
Hi Greg,
For me Hypersepctral just have too much bands to analysed. Unless you have develop a spectral dictionary to makes your job much easier for the classification. The other way to get it done is through enhancement like NDVI or Tasselation Cap but then depends on what and how many classes you want to achive in the output. Object based classification would be best for non organic features since shape can be a easily identified for example squares for buildings and contiguous areas of roads.