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

Patent metadata

Field	Value
Publication number	US-9904849-B2
Application number	US-201715608894-A
Country	US
Kind code	B2
Filing date	May 30, 2017
Priority date	Aug 26, 2015
Publication date	Feb 27, 2018
Grant date	Feb 27, 2018

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 simplified generation of systems for broad area geospatial object detection comprising:
an object model creation module comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operable on the processor, wherein the plurality of programming instructions:
receives a plurality labeled positive training orthorectified geospatial images in at least some of which an object of interest has been identified;
retrieves a plurality of labeled negative training orthorectified geospatial images where objects that are not the object of interest, at least one of which closely resembles the object of interest, have been identified;
programmatically isolates features found in the objects of interest but not in the irrelevant training objects; and
creates at least one object of interest classification model using the features unique to the object of interest;

a machine learning classifier training and verification computer comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operable on the processor, wherein the plurality of programming instructions:
accepts at least one classification model;
retrieves a plurality of labeled and unlabeled orthorectified geospatial training images each comprising the object of interest;
trains a plurality of pre-built machine learning classifier elements, each running a pre-programmed machine learning protocol parameterized with the classification model, using the plurality of labeled and unlabeled orthorectified geospatial training images each comprising the object of interest; and
for each trained machine learning classifier element, verifies performance in classifying the object of interest using a plurality of unlabeled orthorectified geospatial training images comprising the object of interest and a plurality of unlabeled orthorectified geospatial training images that do not contain the object of interest; and

a model-based object classifier comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operable on the processor, wherein the plurality of programming instructions:
retrieve the plurality of trained machine learning elements for the object of interest;
analyzes a plurality of resolution scale-corrected, unanalyzed orthorectified geospatial image segments for presence of at least one object of interest; and
reports the presence and location of any objects of interest found.

2. A method for simplified generation of systems for broad area geospatial object detection the step comprising:
(a) retrieving a plurality of color and spectrally optimized geospatial training images comprising an object of interest that is clearly labeled and a second plurality of color and spectrally optimized geospatial training images that do not contain the object of interest to isolate a set of visual features unique to the object of interest using an pre-engineered object model creation module comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operable on the processor;
(b) employing the set of visual features unique to the object of interest to parameterize at least one pre-engineered machine learning classifier element running at least one pre-programmed machine learning programming protocol using a machine learning classifier element training and verification module comprising a processor, a memory, and a plurality of programming instructions stored in the memory and operable on the processor;
(c) training pre-engineered 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;
(d) refining and confirming the fine specificity of trained machine learning classifier elements for the object of interest to the exclusion of other objects using geospatial training images not containing the object of interest, some of which comprise other irrelevant objects; and
(e) analyzing previously unanalyzed, scale corrected geospatial images for presence of the object of interest using trained machine learning classified element and reporting the results of the study in a format pre-determined by the study author.

Assignees

• Digitalglobe Inc

Inventors

• Estrada Adam
• Green Kevin
• Jenkins Andrew

Classifications

• G06T7/10

  Segmentation; Edge detection (motion-based segmentation G06T7/215) · CPC title

• G06T17/05

  Geographic models · CPC title

• G06T15/50

  Lighting effects · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T7/73

  using feature-based methods · CPC title