Radar Target Recognition by Probabilistic Filtering

摘要

The paper presents a new method of probabilistic filtering for radar target recognition. The classical Bayesian detector/estimator suffers from the insufficient information about target signature probability distributions and their a priory appearance probabilities. If the number of radar image objects to be classified is not known exactly the appeared unknown target may be wrong classified as one of the known targets. To eliminate this type of errors one can use the known probabilistic windows matched by shape to the recognition signature distributions. The combination of the probability window with a non-linear transform of the signature space is proposed in the paper. Such a combination forms a probabilistic filter. The probabilistic filter output is proportional to the likelihood probability of how the sensed object matches to its statistical model. The theoretical background of the probabilistic filtering method and its application to real X-band radar data are presented in the paper. The proposed method reduces the amount of a priory information required for the recognition and detects well the objects of the same nature independently from their size. For example, the probabilistic filter classifies well the different type of vegetation in the radar images.