Although there are many dim target detection and tracking algorithms, they have different adaption to tasks' content and
quality. There is almost no a unified algorithm of target detection and tracking. Reliance on Automated Target
Recognition (ATR) technology is essential to future success of system reasoning and development. In labs, these
algorithms and their combination can be properly evaluated and optimized, and outfield tests and cost may be decreased.
How to analyze and evaluate these algorithms becomes an important problem of ATR. A framework of algorithms'
evaluation has been established. The parameterized simulation method of dim targets has also been proposed. This
method synthesizes simulated targets and the simulated or real background, varying the SNR(Signal Noise Ratio) and
TA(Target Area),etc, and produces quantifiable images. To evaluate the dim target detection or image pretreatment
algorithms for a single target and multiple false alarms, an effective approach SROC (Summary Receiver Operator
Characteristic) based on the ROC (Receiver Operator Characteristic) model has been improved and employed. The
dimension of FAR(False Alarm Rate) has been renewedly defined to adapt to the multiple false alarms. The SROC
model develops and quantifies the ROC model, and obtain a single performance evaluation value, which can better
quantitatively evaluate ATR algorithms. Further, the FROC(Free Receiver Operator Characteristic)model is appropriate
when multiple dim detections are possible and the number of false alarms is unconstrained. The FROC model provides a
qualified method for characterizing both the operational environment and the ability of the ATR algorithm to detect
targets. The FROC model also effectively valuates the detection performance in the situation of single dim target -
multiple false alarms. Tests show the methods are applicable and available in optimizing ATR algorithms and their
combination applications.
|