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 aligning images, comprising:
receiving, at an aligning device, N first type of image sets, wherein N>1, wherein each of the N first type of image sets includes one or more first type of images;
receiving, at the aligning device, N second type of image sets, wherein each of the N second type of image sets includes different one or more second type of images, wherein each of the N second type of image sets is aligned with a different one of the N first type of image sets;
generating, at the aligning device, M transform parameters, wherein M>1;
generating, at the aligning device, M transformed second type of image sets for each of the N first type of image sets so that there are N*M total transformed second type of image sets, wherein each of the M transformed second type of image sets for each of the N first type of image sets is generated by transforming a respective one of the N second type of image sets using a different one of the M transform parameters;
creating, at the aligning device, a regressor configured to identify parameters of a transformation that maps a second type of image set to a first type of image set, wherein the regressor is created based on the N first type of image sets, the M transform parameters, and the N*M total transformed second type of image sets;
receiving, at the aligning device, a target first type of image set and a target second type of image set;
generating, at the aligning device using the regressor, a target transform parameter based on the target first type of image set and the target second type of image set; and
generating, at the aligning device, a transformed target second type of image set by transforming the target second type of image set using the target transform parameter so that the transformed target second type of image set is aligned with the target first type of image set.
2. The method of claim 1 , wherein the first type of images are red-green-blue images, and the second type of images are parcel maps.
3. The method of claim 1 , wherein each image in the first type of image sets, the second type of image sets, the target first type of image set, and the target second type of image set is one of a red-green-blue, panchromatic, infrared, ultraviolet, multi-spectral, or hyperspectral image.
4. The method of claim 1 further comprising determining, at the aligning device, whether the target transform parameter results in convergence.
5. The method of claim 4 further comprising, if it is determined that the target transform parameter does not result in convergence, performing the following steps until convergence: (1) generating, at the aligning device, a new target transform parameter; and (2) determining, at the aligning device, whether the new target transform parameter results in convergence.
6. The method of claim 1 , wherein one or more of the N first type of image sets includes a plurality of co-registered first type of images.
7. The method of claim 1 , wherein the transform parameters and the target transform parameter are for at least one of translation, similarity, perspective, thin-plate-splines, piece-wise affine, B-spline, or high-order bivariate polynomials.