Unsupervised land use and land cover detection
A system and methods for unsupervised land use and land cover detection using a classifier that produces a plurality of class image layers which are filtered to remove misclassified same-label pixel groupings, a class resolution module that reduces multiple pixel labels to a single one if applicable and a reconstruction module that generates the output land use and land cover image.
1. A system for unsupervised land use and land cover detection, comprising:
a spectrum analyzer comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operating on the processor, wherein the programming instructions, when operating on the processor, cause the processor to:
receive an image from an image source; and
algorithmically process the image to produce a plurality of class layers of image information within the image;
a class layer filtering module comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operating on the processor, wherein the programming instructions, when operating on the processor, cause the processor to:
receive a plurality of class layers, wherein the layers comprise at least one of a cloud layer, a soil layer, a vegetation layer, a shadow layer, and a water layer; and
algorithmically process each layer to remove same-class pixel groupings that fail class-related size, shape and color criteria;
hierarchically filter the layers in order from the cloud layer, to the soil layer, to the vegetation layer, to the shadow layer, to the water layer;
a class resolution module comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operating on the processor, wherein the programming instructions, when operating on the processor, cause the processor to:
receive a plurality of filtered class layers;
detect pixels in the stack of class layers that are assigned more than one class; and
perform of plurality of class-assignment confidence comparisons for each multi-class pixel to produce a single class label for that pixel; and
an image reconstruction engine comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operating on the processor, wherein the programming instructions, when operating on the processor, cause the processor to:
receive a plurality of class image layers from the class resolution module; and
reconstruct the colored output image based at least in part on the filtered class image layers.
2. A system according to claim 1 , wherein any dense cloud forms regardless of saturation levels are identified.
3. A system according to claim 1 , wherein rocks, gravel, and sand are excluded.
4. A system according to claim 1 , wherein the cloud layer is detected using a parabolic function fit calculation.
5. A system according to claim 1 , wherein the soil layer is detected using a linear segment fitting calculation.
6. A system according to claim 1 , wherein the vegetation layer is detected using a two-line segment fitting calculation.
7. A system according to claim 1 , wherein the water layer is detected using a normalized difference water index ratio discrimination calculation.