Systems and methods for analyzing remote sensing imagery
Disclosed systems and methods relate to remote sensing, deep learning, and object detection. Some embodiments relate to machine learning for object detection, which includes, for example, identifying a class of pixel in a target image and generating a label image based on a parameter set. Other embodiments relate to machine learning for geometry extraction, which includes, for example, determining heights of one or more regions in a target image and determining a geometric object property in a target image. Yet other embodiments relate to machine learning for alignment, which includes, for example, aligning images via direct or indirect estimation of transformation parameters.
1. A method of generating a label image based on a parameter set, comprising:
receiving, at an object detector, training images;
receiving, at the object detector, training parameter sets, wherein each of the training parameter sets corresponds to a transformed representation of one or more labels, and wherein the one or more labels correspond to a different training image;
creating, at the object detector, a classifier configured to determine a parameter set for an image based on the training images and the training parameter sets;
receiving, at the object detector, a target image;
determining, at the object detector using the classifier, a target parameter set that corresponds to one or more target labels for the target image, wherein the target parameter set corresponds to a target transformed representation of the one or more target labels; and
generating, at the object detector, a label image by applying an inverse target parameter set to the target image, wherein the inverse target parameter set corresponds to an inverse transformation of a transformation represented by the target parameter set, wherein one or more pixels of the label image are associated with a class.
2. The method of claim 1 , wherein each of the one or more classes is either two-class or multi-class.
3. The method of claim 1 further comprising receiving second training parameter sets for the training images, wherein the second training parameter sets comprise at least one of time, date, sun direction, sun position, latitude, or longitude.
4. The method of claim 1 , wherein each of the training parameters sets and the target parameter set represents a transformation created using one of discrete cosine transform, wavelets, discrete Fourier transform, principal component analysis, non-negative Matrix Factorization, or Hadamard transform.
5. The method of claim 1 , wherein each of the training images and the target image is one of a red-green-blue, panchromatic, infrared, ultraviolet, multi-spectral, or hyperspectral image.