Raytheon has extensively processed high-resolution sidescan sonar images with its Automatic Target Recognition
(ATR) algorithms to classify target-like objects (TLOs) in a variety of underwater environments. The ATR algorithm
segments the image into candidate highlight and shadow regions of interest (ROIs), and extracts and scores features
from these ROIs. The TLOs are classified by thresholding an overall classification score, formed by summing the
individual feature scores. The algorithm performs reliably against TLOs that exhibit highlight and shadow regions that
are both distinct relative to the ambient background. However, the sonar images for many real-world undersea
environments can contain a significant percentage of TLOs exhibiting either "weak" highlight or shadow regions.
Robust performance in these environments is achieved by tailoring the individual feature scoring algorithms to optimize
the separation between the corresponding highlight or shadow feature scores of targets and non-targets. This study
examines modifications to a previously presented alternate approach that employs Fisher fusion principles to generate
optimal weighting coefficients which are applied to the individual feature scores before final classification processing.
Results from processing of at-sea data sets demonstrate the performance benefits obtained from the modifications.
Raytheon has extensively processed high-resolution sidescan sonar images with its CAD/CAC algorithms to provide
classification of targets in a variety of shallow underwater environments. The Raytheon CAD/CAC algorithm is based
on non-linear image segmentation into highlight, shadow, and background regions, followed by extraction, association,
and scoring of features from candidate highlight and shadow regions of interest (ROIs). The targets are classified by
thresholding an overall classification score, which is formed by summing the individual feature scores. The algorithm
performance is measured in terms of probability of correct classification as a function of false alarm rate, and is
determined by both the choice of classification features and the manner in which the classifier rates and combines these
features to form its overall score. In general, the algorithm performs very reliably against targets that exhibit "strong"
highlight and shadow regions in the sonar image- i.e., both the highlight echo and its associated shadow region from the
target are distinct relative to the ambient background. However, many real-world undersea environments can produce
sonar images in which a significant percentage of the targets exhibit either "weak" highlight or shadow regions in the
sonar image. The challenge of achieving robust performance in these environments has traditionally been addressed by
modifying the individual feature scoring algorithms to optimize the separation between the corresponding highlight or
shadow feature scores of targets and non-targets. This study examines an alternate approach that employs principles of
Fisher fusion to determine a set of optimal weighting coefficients that are applied to the individual feature scores before
summing to form the overall classification score. The results demonstrate improved performance of the CAD/CAC
algorithm on at-sea data sets.
Raytheon has extensively processed high-resolution sonar images with its CAD/CAC algorithms to provide real-time
classification of mine-like bottom objects in a wide range of shallow-water environments. The algorithm performance
is measured in terms of probability of correct classification (Pcc) as a function of false alarm rate, and is impacted by
variables associated with both the physics of the problem and the signal processing design choices. Some examples of
prominent variables pertaining to the choices of signal processing parameters are image resolution (i.e., pixel
dimensions), image normalization scheme, and pixel intensity quantization level (i.e., number of bits used to represent
the intensity of each image pixel). Improvements in image resolution associated with the technology transition from
sidescan to synthetic aperture sonars have prompted the use of image decimation algorithms to reduce the number of
pixels per image that are processed by the CAD/CAC algorithms, in order to meet real-time processor throughput
requirements. Additional improvements in digital signal processing hardware have also facilitated the use of an
increased quantization level in converting the image data from analog to digital format. This study evaluates
modifications to the normalization algorithm and image pixel quantization level within the image processing prior to
CAD/CAC processing, and examines their impact on the resulting CAD/CAC algorithm performance. The study
utilizes a set of at-sea data from multiple test exercises in varying shallow water environments.
KEYWORDS: Image processing, Detection and tracking algorithms, Image fusion, Data fusion, Image classification, Image segmentation, Computer aided diagnosis and therapy, Land mines, Classification systems, Image processing algorithms and systems
Over the past several years, Raytheon Company has adapted its Computer Aided Detection/Computer-Aided Classification (CAD/CAC) algorithm to process side-scan sonar imagery taken in both the Very Shallow Water (VSW) and Shallow Water (SW) operating environments. This paper describes the further adaptation of this CAD/CAC algorithm to process Synthetic Aperture Sonar (SAS) image data taken by an Autonomous Underwater Vehicle (AUV). The tuning of the CAD/CAC algorithm for the vehicle's sonar is described, the resulting classifier performance is presented, and the fusion of the classifier outputs with those of another CAD/CAC processor is evaluated. The fusion algorithm accepts the classification confidence levels and associated contact locations from the different CAD/CAC algorithms, clusters the contacts based on the distance between their locations, and then declares a valid target when a clustered contact passes a prescribed fusion criterion. Three different fusion criteria are evaluated: the first based on thresholding the sum of the confidence factors for the clustered contacts, the second based on simple binary combinations of the multiple CAD/CAC processor outputs, and the third based on the Fisher Discriminant. The resulting performance of the three fusion algorithms is compared, and the overall performance benefit of a significant reduction of false alarms at high correct classification probabilities is quantified.
In 1999 Raytheon adapted its shallow-water Side-Looking Sonar (SLS) CAD/CAC algorithm to process side-scan sonar data obtained with the Woods Hole Oceanographic Institute's Remote Environmental Monitoring Units (REMUS) autonomous underwater vehicle (AUV). To date, Raytheon has demonstrated the ability to effectively execute mine-hunting missions with the REMUS vehicle through the fusion of its CAD/CAC algorithm with several other CAD/CAC algorithms to achieve a high probability of correct classification while maintaining a low false alarm rate. Raytheon recently reported CAD/CAC algorithm enhancements that demonstrated a significant improvement in overall CAD/CAC performance across a diverse set of environments. Additional algorithm enhancements that further improve performance over this same set of environments are described herein. The paper also presents results obtained from processing this diverse environmental data set with the enhanced Raytheon CAD/CAC algorithm, and the performance achieved by fusing the Raytheon CAD/CAC outputs with those of the other CAD/CAC algorithms.
In 1999 Raytheon adapted its shallow-water Side-Looking Sonar (SLS) Computer Aided Detection/Computer Aided Classification (CAD/CAC) algorithm to process side-scan sonar data obtained with the Woods Hole Oceanographic Institute's Remote Environmental Monitoring Units (REMUS) autonomous underwater vehicle (AUV). To date, Raytheon has demonstrated the ability to effectively execute mine-hunting missions with the REMUS vehicle through the fusion of its CAD/CAC algorithm with several other CAD/CAC algorithms to achieve low false alarm rates while maintaining a high probability of correct detection/classification. Mine-hunting in the very shallow water (VSW) environment poses a host of difficulties including such issues as: a higher incidence of man made clutter, significant interference due to biological sources (such as kelp or silt), the scouring of mines into the bottom, interference from surface/bottom bounce, and image distortion due to vehicle motion during image generation. These issues coupled with highly variable bottom conditions and small bottom targets make reliable hunting in the VSW environment very difficult. In order to be operationally viable, the individual CAD/CAC algorithms must demonstrate robustness over these very different mine-hunting environments. A higher normalized false alarm rate per algorithm is considered acceptable based on the false alarm reduction achieved through multi-algorithm fusion. Raytheon's recent CAD/CAC algorithm enhancements demonstrate a significant improvement in overall CAD/CAC performance across a diverse set of environments, from the relatively benign Gulf of Mexico environment to the more challenging areas off the coast of southern California containing significant biological and bottom clutter. The improvements are attributed to incorporating an image normalizer into the algorithm's pre-processing stage in conjunction with several other modifications. The algorithm enhancements resulted in an 11% increase in overall correct classification probability with an accompanying 17% reduction in false alarm rate, when averaged over the multiple environments. The paper discusses the algorithm enhancements and presents the detailed performance results.
Over the past several years, Raytheon Company has adapted its Computer Aided Detection/Computer-Aided Classification (CAD/CAC)algorithm to process side-scan sonar imagery taken in both the Very Shallow Water (VSW) and Shallow Water (SW) operating environments. This paper describes the further adaptation of this CAD/CAC algorithm to process SW side-scan image data taken by the Battle Space Preparation Autonomous Underwater Vehicle (BPAUV), a vehicle made by Bluefin Robotics. The tuning of the CAD/CAC algorithm for the vehicle's sonar is described, the resulting classifier performance is presented, and the fusion of the classifier outputs with those of three other CAD/CAC processors is evaluated. The fusion algorithm accepts the classification confidence levels and associated contact locations from the four different CAD/CAC algorithms, clusters the contacts based on the distance between their locations, and then declares a valid target when a clustered contact passes a prescribed fusion criterion. Four different fusion criteria are evaluated: the first based on thresholding the sum of the confidence factors for the clustered contacts, the second and third based on simple and constrained binary combinations of the multiple CAD/CAC processor outputs, and the fourth based on the Fisher Discriminant. The resulting performance of the four fusion algorithms is compared, and the overall performance benefit of a significant reduction of false alarms at high correct classification probabilities is quantified. The optimal Fisher fusion algorithm yields a 90% probability of correct classification at a false alarm probability of 0.0062 false alarms per image per side, a 34:1 reduction in false alarms relative to the best performing single
CAD/CAC algorithm.
The fusion of multiple Computer Aided Detection/Computer Aided Classification (CAD/CAC) algorithms has been shown to be effective in reducing the false alarm rate associated with the automated classification of bottom mine-like objects when applied to side-scan sonar images taken in Very Shallow Water (VSW) environments. This paper reports on the application of such CAD/CAC Fusion algorithms to the shallow water environment, using sidescan sonar data taken in the Gulf of Mexico during April 2000. The fusion algorithm accepts the classification confidence levels and associated contact locations from two different CAD/CAC algorithms, clusters the contacts based on the distance between their locations, and then declares a valid target when a clustered contact passes a prescribed fusion criterion. Two different fusion criteria are evaluated: the first based on the Fisher Discriminant, and the second based on a constrained optimization approach, which minimizes the total number of false alarms over the clustering distance and cluster confidence factor thresholds for a given probability of correct classification. The Fisher-based fusion provided an 82% probability of correct classification at a false alarm rate of 0.034 false alarms per image per side (port or starboard). This performance represented a 2:1 reduction in false alarms over a single CAD/CAC algorithm at this same probability of correct classification. The cluster confidence fusion algorithm performed nearly as well, yielding the 82% correct classification probability at a false alarm rate of 0.039 false alarms per image per side.
The performance of Computer Aided Detection/Computer Aided Classification (CAD/CAC) Fusion algorithms on side-scan sonar images was evaluated using data taken at the Navy's's Fleet Battle Exercise-Hotel held in Panama City, Florida, in August 2000. A 2-of-3 binary fusion algorithm is shown to provide robust performance. The algorithm accepts the classification decisions and associated contact locations form three different CAD/CAC algorithms, clusters the contacts based on Euclidian distance, and then declares a valid target when a clustered contact is declared by at least 2 of the 3 individual algorithms. This simple binary fusion provided a 96 percent probability of correct classification at a false alarm rate of 0.14 false alarms per image per side. The performance represented a 3.8:1 reduction in false alarms over the best performing single CAD/CAC algorithm, with no loss in probability of correct classification.
A method for combining the outputs of three different computer aided detection/computer aided classification (CAD/CAC) algorithms is presented and applied to a set of sidescan sonar data taken in the very shallow water environment, where the CAD/CAC algorithms are each tuned to detect mine-like objects. The fusion center receives from each algorithm the planar image coordinates and a confidence factor associated with individual CAD/CAC contacts, and produces fused classification reports of the mine-like objects. Since the three CAD/CAC algorithms use very different approaches, we make the reasonable assumption that valid classifications are nearby each other and false alarms occur randomly in the image. The resultant geometric clustering eliminates most of the false alarms while maintaining a high level of correct classification performance. Our unique fusion algorithm takes a constrained optimization approach, which minimizes the total number of false alarms over the clustering distance and cluster confidence factor thresholds for a given probability of correct classification. Resultant receiver operating characteristics show a significant reduction in the number of false contacts: the false alarm rate from any individual CAD/CAC algorithm is reduced by a factor of four or greater through the optimized data fusion processing.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.