Performance Evaluation for Radar Signal Recognition Based on AHP - Multimedia and Signal Processing - page 555

2 Indices for the Recognition Performance Evaluation

Evaluation indices selection is determined by the demand of performance evaluation,
based on the focus concerned performance, form indices that can be quantified.

2.1 Meaning and Calculation of the Recognition Rate

The comparison of radar signals recognition performance is generally based on the
recognition rate, namely in the recognition test. The pulse number of correct recogni-
tion divided by the total number of pulses is the signal recognition rate of the
condition, which can also be interpreted as the recognition ability of the method to the
target signal in the condition. This understanding is correctly identified as "1" on the
premise that it is falsely or not identified as "0". Supposing the true value of the rec-
ognition rate of the method is p , thus the correct probability of the i th recognition is
p .The correct probability of the i th recognition is as the law of 01

distribution.
Do Bernoulli (Bernoulli) experiment of signal recognition, then the number of correct
recognition obeys the distribution of

−

Bnp , n is the number of experiments.

(, )

2.2 Measurement of Recognition Rate

For the general sense, solving the recognition rate is the result of limited experiments,
which can only be infinitely close to the true value and can not achieve the true value.
If for different tests, it can be found that the recognition rate is fluctuant variable, and
the variation range as well as the number of great changes of the recognition rate re-
flects the recognition performance.
The paper introduces the concept of measurement of recognition rate [6] (MRR),
dividing n test results into m subgroups, which can separately acquire the mean
value of each group of nm test results, then we get m MRR samples. For
radar signals recognition, it is supposed that there are l radar radiating sources
which will eventually generate lm

×

MRR samples. Because of the consistency of
the algorithm and solution, it is only needed to research MRR samples of a single
radar radiation source, namely m MRR samples. Because MRR is a variable which
has distribution, mean and variance, together with the independence of external
conditions, by using its characters, we can strictly obtain accurate estimation of radar
signals recognition results.

2.3 Acquirement of MRR Samples

Every MRR sample, which contains distribu-

tion from the above analysis. Based on independence and identical distribution as well
as the central limit theorem, MRR obeys normal distribution which can be proven[7].
Literature [8] thinks that M is bigger than 50 meets the large sample capacity of
using the central limit theorem of independence and same distribution. When

nm recognition results obeys 01 −