Observation is the foundation of any science. But understanding, analysing, processing and transforming it into data that transform everyday life is another history.
Human activities have reached a complexity level that is no longer possible to evaluate just from the ground. Satellite imagery is not a human eye, but it started to take a significant role in clearing the perspective of this complexity.
This vast amount of growing data requires machines as well as humans. Companies and different types of organizations can now access this data privileged to few just until recently, due to data democratization.
Geospatial intelligence is about this data: access it and use it for economic, logistic or other humanitarian activities.
Content based image retrieval (CBIR) is a process framework for efficiently retrieving images from a collection by similarity.
The query relies on extracting the appropriate characteristics quantities describing the desired contents of images, not on keywords, tags or description associated with image.
Well, we know it’s mind blowing, but there is a good logic…you’ll see!
Geospatial Intelligence is applying the CBIR technique to data acquired via Earth Observation (EO) technologies (satellites, remote sensing), and it is a general-purpose image collections search engine.
The service finds similar patches over large amounts of satellite images data, according to your query.
How the service works:
You define your area of interest — for example, river/road in a city, vessel containers, storage tanks, oilfields etc — and the service finds similar patches over large amounts of data.
Each image patch is characterized based on specific descriptors adapted to EO image particularities — colours, shapes, texture, or any other information that can be derived from the image itself.
The algorithm searches for content similarities and delivers several visualization approaches from which you can narrow your findings.
Search can be done both in the selected scene or in data base, and it is using optic and radar scenes.
Active learning — an opportunity afforded by machine learning
You can coach the service, by applying positive and negative examples. The application will assign a specific class, while the machine learning algorithms will perform the semantic annotation.
Several visualization approaches of the search results will be delivered, highlighting the patches or the scenes most similar to your specific query area.
Every industry or sector willing to address its activities from a larger, more comprehensible, but yet approachable perspective can benefit:
- industrial, transport and logistic activities;
- oil & gas, energy and natural resources;
- public authorities;
- managing living resources.
We live in complex economic and social environments, but data can give us information about where is the problem, and also how to address it…without blowing our mind away.
The service was first developed for ESA project — Open Source Image Retrieval — Integration of Developed Tools (OSIRIDE), together with University Politehnica of Bucharest — Research Center for Spatial Information (UPB — CeoSpaceTech).