# US 2021/0342585 A1 · Systems and methods for extracting and vectorizing features of satellite imagery

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2021/0342585 A1
- Document ID: US20210342585A1
- Status: Published application
- Kind: A1
- Application number: 16/864,756
- Filed: 2020-05-01
- Published: 2021-11-04
- Earliest claimed: 2020-05-01
- Assignees: CACI, Inc.—Federal
- Applicants: CACI, Inc. - Federal
- Inventors: Jacob A. Fleisig; Evan M. Colvin; Peter Storm Simonson; Nicholas Grant Chidsey
- Web page: https://patentgrep.com/patent/US20210342585A1

## CPC

- G06T 7/12: Edge-based segmentation
- G06V 20/176: {Urban or other man-made structures}
- G01C 21/3815: {Road data}
- G01C 21/3852: {Data derived from aerial or satellite images}
- G06F 17/16: Matrix or vector computation {, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization (matrix transposition G06F7/78)}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G06T 5/70: Denoising; Smoothing
- G06T 7/155: involving morphological operators
- G06T 17/05: Geographic models
- G06V 10/34: Smoothing or thinning of the pattern; Morphological operations; Skeletonisation
- G06V 10/457: {by analysing connectivity, e.g. edge linking, connected component analysis or slices}
- G06V 10/762: using clustering, e.g. of similar faces in social networks
- G06V 20/10: Terrestrial scenes (scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59)
- G06V 20/182: {Network patterns, e.g. roads or rivers}
- G06N 3/048: Activation functions
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06T 2207/10032: Satellite or aerial image; Remote sensing
- G06T 2207/20044: Skeletonization; Medial axis transform
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30184: Infrastructure
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G01C: MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY (measuring liquid level G01F; radio navigation, determining distance or velocity by use of propagation effects, e.g. Doppler effects, propagation time, of radio waves, analogous arrangements using other waves G01S)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system may be configured to collect geospatial features (in vector form) such that a software application is operable to edit an object represented by at least one vector. Some embodiments may: generate, via a trained machine learning model, a pixel map based on an aerial or satellite image; convert the pixel map into vector form; and store the vectors. This conversion may include a raster phase and a vector phase. A system may be configured to obtain another image, generate another pixel map based on the other image, convert the other pixel map into vector form, and compare the vectors to identify changes between the images. Some implementations may cause identification, based on a similarity with converted vectors, of a more trustworthy set of vectors for subsequent data source conflation.

## Claims

### Claim 1

A computer-implemented method, comprising:
generating, via a machine learner, a pixel map;
converting the pixel map into a plurality of first vectors; and
storing, in a file type, the first vectors such that a software application is operable to edit a first object represented by at least one of the first vectors.

### Claim 2 (depends on 1)

The method of claim 1, wherein the conversion comprises a raster phase, which includes converting the pixel map into skeleton vectors by:
performing morphological-cleanup, which comprises performing erosion to remove noise and artifacts at a pixel level; and
performing skeletonization, which comprises thinning the pixel map to a thickness of one pixel.

### Claim 3 (depends on 2)

The method of claim 2, further comprising:
performing vectorization, which comprises converting the thinned pixel map to vector form by:
traversing a skeleton map to find one or more neighboring pixels, which represent one or more second objects of a same type as the first object, within a predetermined pixel distance;
extracting a vector in a direction of each of the one or more found pixels; and
combining the one or more extracted vectors.

### Claim 4 (depends on 3)

The method of claim 3, further comprising:
performing smoothing, which comprises removing a crookedness of the combination.

### Claim 5 (depends on 2)

The method of claim 2, wherein the conversion further comprises a vector phase, which includes improving quality and shape of second vectors of the pixel map and removing artifacts by:
performing connectivity-graphing, which includes adding the second vectors to a connectivity graph; and
performing gap-jumping, which includes detecting gaps at dead-ends in the connectivity graph.

### Claim 6 (depends on 1)

The method of claim 1, further comprising:
performing cluster-collapsing by:
clustering a plurality of nodes that have distances from each other that each satisfies a criterion; and
joining objects that are not connected to each other and that have distances from each other that each satisfies another criterion.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
performing spur-removal, which includes:
detecting one or more spurs at a dead-end of a line; and
removing the one or more spurs, which are shorter than a threshold length.

### Claim 8 (depends on 5)

The method of claim 5, further comprising:
performing intersection-repair, which includes:
repairing one or more four-way intersections by collapsing detected instances of two three-way intersections; and
repairing one or more three-way intersections, including when one object intersects but does not pass through another object of the same type, by utilizing the connectivity graph to find distorted intersections and by ignoring an area near the distorted intersection such that:
two lines are colinear; and
the one or more repaired intersections is placed where a colinear pair and third objects meet.

### Claim 9 (depends on 8)

The method of claim 8, further comprising:
performing vertex reduction, which includes processing the lines to reduce vertex count.

### Claim 10 (depends on 1)

The method of claim 1, further comprising:
determining whether each pixel of the map belongs to an object of a same type as the first object.

### Claim 11 (depends on 4)

The method of claim 4, wherein the generation comprises outputting a plurality of pixels associated with one or more intermediate values that do not satisfy a criterion for indicating whether each of the plurality of pixels belongs to the object of the same type as the first object.

### Claim 12 (depends on 5)

The method of claim 5, further comprising:
adjusting a value of each pixel associated with one of the one or more intermediate values such that the each pixel is determined to belong to the object of the same type as the first object or determined not to belong to the object of the same type as the first object.

### Claim 13 (depends on 1)

The method of claim 1, further comprising:
transforming the first vectors from a pixel-space to a coordinate system,
wherein the transformed vectors are stored.

### Claim 14 (depends on 10)

The method of claim 10, wherein the same type is a road or building.

### Claim 15

A method, comprising:
obtaining, via one or more processors, first and second aerial images, wherein the first and second images are of a same area and are taken at different times;
generating, via a machine learner, first and second pixel maps using the first and second images, respectively;
converting, via one or more processors, the first and second pixel maps into a plurality of first vectors and a plurality of second vectors, respectively; and
comparing the pluralities of first and second vectors to identify one or more changes.

### Claim 16 (depends on 15)

The method of claim 15, wherein each of the first and second pixel maps is generated using the respective image.

### Claim 17 (depends on 15)

The method of claim 15, wherein the machine learner implements a convolutional neural network (CNN).

### Claim 18

A method, comprising:
obtaining, via one or more processors, an image;
generating, via a machine learner, a pixel map using the image;
converting, via one or more processors, the pixel map into a plurality of first vectors;
obtaining, via one or more processors, a plurality of second vectors and a plurality of third vectors;
comparing the pluralities of first and second vectors to determine a first similarity level;
comparing the pluralities of first and third vectors to determine a second similarity level; and
identifying the plurality of second vectors or the plurality of third vectors as being more accurate based on the first and second similarity levels.

### Claim 19 (depends on 18)

The method of claim 18, wherein the conversion comprises a raster phase.

### Claim 20 (depends on 19)

The method of claim 19, wherein the conversion further comprises a vector phase.
