Advanced Target Recognition

Thumbnail: click to enlarge

SMART™ Methodology

KBSI has recently been awarded a Phase I contract to address the shortcomings of current ship detection and identification systems for the U.S. Navy.  The System for MARitime Targets Recognition and Retrieval (SMART™) technology, in contrast to current systems that focus on image processing techniques only, will integrate image classification, data simulation, and data fusion technologies and techniques.  The SMART™ technology will have the capability to synthesize a vessel model of the target of interest from existing information, to simulate Infrared Sea Surface Skin Temperature Radiometer (ISAR) or other image sources based on the constructed model and selected environmental factors, and to generate initial match profiles from this synthetic source. 

Current maritime surveillance systems face two significant challenges. First, there are a limitless number of vessels/vessel configurations that can be used for military purposes, and each vessel can exhibit different characteristics under different configurations/conditions (for example, commercial vessels can change their characteristics significantly by simply loading or unloading cargo).  To establish a database that stores all these vessels under various conditions is impossible.  Second, the methods for data acquisition, while providing advantages, also each have significant limitations. ISAR radar can operate at a long distance from the target and can operate in both day and night.  The quality of ISAR images, however, depends on environmental factors (e.g., sea-state) and the dimension and movements of the subject vessel, and even the best ISAR images cannot provide detailed vessel features. Infrared (IR) images, while providing higher resolution than ISAR images, must be acquired at a relatively close range and are highly sensitive to the environmental conditions as well. High quality photographs can provide fine details of a vessel, but the quality of these images depend upon the illumination available.  An effective surveillance system must make the best use of all available images and information, integrating and fusing them to overcome the shortcomings with which the individual components are plagued. 

In the SMART™ technology's image classification scheme, each type of image—ISAR, IR, or photograph—is processed by an independent SMART™ technology component.  In the case of an ISAR image, for example, the SMART™ technology uses a vessel model constructor and an ISAR image simulator. The vessel models are digital models that represent the key geometric features of a vessel and that are used as templates in the SMART™ technology . The SMART™ technology generates the target of interest model by configuring a template model similar to the target and by taking all available information into account. For example, a photograph may provide detailed geometric features differentiating the target from other vessels, and intelligence may provide information concerning any recent changes to the target configuration (i.e., newly loaded cargo). Based on the vessel model and environmental factors (i.e., sea-state), the SMART™ technology's ISAR image simulator generates reference ISAR images. The acquired ISAR image is compared with the reference ISAR images by the ISAR image comparator. Each image (ISAR, infrared, photo), when it is used to construct the digital vessel model, has a set of matching points within the vessel model. Thus, models from different images can be cross-indexed between disparate catalogs and matched.  Based on the similarity between the acquired and the reference images and the classification result produced by the ISAR image classifier, the fusion engine makes the final decision of the target. 

The SMART™ technology will greatly improve the Navy’s surveillance capabilities in the short term.  In addition, the SMART™ technology enables a continuously improving match dataset of targets of interest—a significant advance over current surveillance systems.  SMART™ constructs an initial synthetic model of the target of interest using available information (i.e., type of target, photos, line drawings, etc.), and then generates simulated prospective recognition points across the various catalogs within the match database. This approach facilitates incremental improvement in the match database:  once the target is imaged during surveillance, the database can be updated from this real data.  The SMART™ technology approach will provide increased surveillance proficiency and increased security.