Radiometric identification of emitters has been attracting attention to reinforce security of wireless systems. It is expected to distinguish subtle differences among emitters even of the same model based only on the measured signals to prevent a fraudulent device from accessing a system. This paper presents identification of emitters used in the Automatic Identification System, which is regarded as a key part for achieving maritime awareness. Physically measured signals show the characteristics of subtle differences among emitters. Based on these characteristics two novel methods for identification are investigated, the optimal linear discriminant function is derived from the optimal Bayesian estimator to make the best of the limited number of samples. A difference vector of a periodic signal over the period is also introduced to extract subtle difference among signals. Their cumulants of higher orders, which are invariant statistics against additive Gaussian noise in principle, are investigated to be classified by probabilistically integrated support vector classifiers. Their applications to signals measured through wire from emitters of the same model as well as through an antenna in field from emitters of assumed same model show promising results for radiometric identification of emitters.