Triangle identification algorithm has been widely applied in star pattern identification of star sensor. However, its lower characteristic dimensions lead to matching redundancy and identification errors. To improve the identification accuracy, the identification process of the triangle identification algorithm is improved. First, the method of ''quasi uniform distribution'' of space solid angles is proposed. That is, based on the size of space solid angle a certain zone at the celestial equator corresponds to, the whole celestial sphere is partitioned into a number of sub-blocks. Then, guide stars are selected uniformly according to sub-blocks, obtaining a guide star catalogue with fewer stars, better distribution uniformity and higher completeness. Finally, an approach is proposed to further optimize primarily selected observation triangle groups for the second time for matching identification. The constraint relationship between the sides and included angles of the observation triangle is converted onto the image plane. The simulation experiment result shows that the star pattern identification technique based on the improved triangle algorithm has strong anti-interference performance against star point position noises, magnitude (brightness) noises and ''missing stars'' and it is superior to traditional algorithms in identification rate.