Skip Navigation LinksATR

Du er her: Skip Navigation Linksffi.no / no / Forskningen / Avdeling Maritime systemer / HUGIN / Automatic Target Recognition (ATR)
Forskningen
Automatic Target Recognition (ATR)
Skriv ut
Tips venn
Facebook
Twitter
02.02.2011
Linje

What is ATR?​

Automatic Target Recognition (ATR) is the process where computer algorithms detect and classify specified types of military targets or objects in sensor data. The sensor data is typically high-resolution, large-area images from SAR, SAS, LADAR, forward looking infra-red (FLIR), video, etc.
ATR is a multi-disciplinary field that requires expertise in sensor technology, signal/image processing, pattern recognition, computer programming and system architecture. The main challenge for ATR is to simultaneously achieve:
  • High target recognition accuracy (recognize targets as correct class) in spite of the combinatorial explosion of target signature variations due to target configuration, target/sensor acquisition parameters, target/background interaction and environment variations
  • Low false alarm rates (avoid recognizing
    non-targets as targets) even in varying and complex backgrounds
  • Real time operation
 
Another important challenge for ATR is the evaluation and prediction of ATR field performance given the practical limitation of training data sets that cannot represent the extreme variability of the real world.

 

ATR architecture

The architecture of a basic ATR system is shown to the right. The detection module searches the complete sensor data for target-like signatures. A vector of feature values is calculated at each detected location. The features are related to response size, shape, texture and/or signal statistics, and have been selected during algorithm training for their abilities to distinguish between the various target classes and background. The classifier receives the feature vector and uses statistical methods to find the most probable corresponding target class. Image enhancement such as noise suppression can be applied both before the detection and feature extraction stages.

 

Why ATR?

ATR has become a vital capability for increased effectiveness and efficiency of warfare systems operating from a variety of platforms for targeting and surveillance. The reason for this is twofold:

Firstly, the immense data rate of modern imaging sensors combined with strict time constraints in military operations often prohibits human analysis of all collected data. Automated processing is thus required, at least as data reduction tools to select regions of interest out of a wide-area scene. Only data from these regions are then evaluated further. Detailed analysis of sensor data over extended periods is challenging for humans, and ATR can also be used to support inexperienced or fatigued operators. The processing is then computer aided, instead of fully automated.
Secondly, automatic analysis of sensor data is required to realize autonomous vehicles that are capable of adaptive behaviour based on real-time perception of their surroundings. Such intelligent, mobile robots have until recently been confined to the sci-fi literature, but are expected to play increasingly important military roles both on land, in air and underwater. When only narrow-band communication links are available, most sensor data processing must be performed onboard the vehicle. Examples of adaptive behaviour are selective transmission of sensor data based on ATR results and in-mission revisiting of recognised targets for additional observations.
 

ATR for HUGIN AUV

Over the last decade, FFI has developed ATR technology for HUGIN AUV using data from various sonar types and sea environments. The development has been focused on recognition of seafloor mines, but also includes other applications such as pipeline inspection.
The first at-sea usage of an experimental algorithm was successfully performed during the first HUGIN operation from a Royal Norwegian Navy (RNoN) minehunter in December 2001. The first ATR system designed for operative duty was delivered to the RNoN in 2008-9 as part of the HUGIN 1000-MR delivery. The system is based on data from the Kongsberg Maritime HISAS 1030 synthetic aperture sonar (SAS) and is aimed at bottom mines in benign environments.
The architecture of a basic ATR system. Illustration: FFI
© Copyright Forsvarets Forskningsinstitutt