System for simplified generation of systems for broad area geospatial object detection

Patent metadata

Field	Value
Publication number	US-11462007-B2
Application number	US-202017069776-A
Country	US
Kind code	B2
Filing date	Oct 13, 2020
Priority date	Aug 26, 2015
Publication date	Oct 4, 2022
Grant date	Oct 4, 2022

Abstract

A system for simplified generation of systems for analysis of satellite images to geolocate one or more objects of interest. A plurality of training images labeled for a study object or objects with irrelevant features loaded into a preexisting feature identification subsystem causes automated generation of models for the study object. This model is used to parameterize pre-engineered machine learning elements that are running a preprogrammed machine learning protocol. Training images with the study are used to train object recognition filters. This filter is used to identify the study object in unanalyzed images. The system reports results in a requestor's preferred format.

First claim

What is claimed is:

1. A system for broad area geospatial object detection comprising:
at least one computing device comprising a processor, a memory, a network interface, and a plurality of programming instructions stored in the memory and operable on the processor;
an object model creation module comprising programming instructions operating on the processor of one of the computing devices to cause the respective processor to:
receive a plurality of orthorectified geospatial images in which an object of interest has been identified;
retrieve a plurality of orthorectified geospatial images wherein objects that are not the object of interest have been identified; and
train an object classification model to classify only the object of interest;

a machine learning classifier training and verification module comprising programming instructions operating on the processor of one of the computing devices to cause the respective processor to:
accept the object classification model;
retrieve a plurality of labeled and unlabeled orthorectified geospatial training images each containing at least one instance of the object of interest; and
train a plurality of machine learning classifier elements, each running a machine learning protocol parameterized with the object classification model, using the plurality of labeled and unlabeled orthorectified geospatial training images; and

a model-based object classifier comprising programming instructions operating on the processor of one of the computing devices to cause the respective processor to:
retrieve the plurality of trained machine learning elements for the object of interest;
analyze a plurality of resolution scale-corrected, unanalyzed orthorectified geospatial image segments for presence of at least one object of interest; and
report the presence and location of any objects of interest found.

2. A method for broad area geospatial object detection, the method comprising the steps of:
retrieving a first plurality of geospatial training images each containing at least one labeled instance of an object of interest, and a second plurality of geospatial training images that do not contain the object of interest;
isolating a set of visual features unique to the object of interest using an object model creation module;
employing the set of visual features unique to the object of interest to parameterize at least one machine learning classifier running at least one machine learning protocol, using a machine learning classifier element training and verification module;
training the machine learning classifier elements to identify the object of interest using a plurality of training geospatial images with the object of interest labeled in one subset and not labeled in a second subset within the machine learning classifier element training and verification module;
analyzing previously unanalyzed geospatial images for presence of the object of interest using the trained machine learning classified elements; and
reporting the presence and location of any objects of interest found.

Assignees

• Digitalglobe Inc

Inventors

• Estrada Adam
• Green Kevin
• Jenkins Andrew

Classifications

• G06V10/7753

  Incorporation of unlabelled data, e.g. multiple instance learning [MIL] · CPC title

• G06V10/774

  Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06V20/13 Primary

  Satellite images · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title