Abstract: Triangle algorithm is one of the most classical and widely used method of star pattern identification, but it has short comings of large search range, redundancy match, weak anti-noise ability, and so on. Applying the neural network technology to the satellite image recognition, a new recognition method combined with the excellent classification ability of neural network technology and the reliable angular matching ability of the triangle algorithm was proposed. The feature vector sofeachguidestar constructed based on the distribution of the neighboring stars were used to train the SOM network of the star pattern recognition system so that the classification function was obtained. The network could identify the class of the unknown star by the feature vector. In the corresponding triangle database, the triangle algorithm was applied to find the matching triangle and identify the star map. Results of simulated identification show that the new method can reduce the star catalog, identify the star map fast and make the system more robust with respect to noise. In the whole sky identification, the average recognition time is less than 5ms, and the recognition rate is still as high as 99% when the noise standard deviation is 0.025.

Key words: tar sensor, star pattern identification, neural network, self-organizing map network, triangle algorithm