Aircraft identification is implemented on thermal images acquired from ground-to-air infrared cameras. SRC is proved to be an effective image classifier robust to noise, which is quite suitable for thermal image tasks. However, rotation invariance is challenging requirements in this task. To solve this issue, a method is proposed to compute the target main orientation firstly, then rotate the target to a reference direction. Secondly, an over-complete dictionary is learned from histogram of oriented gradient features of these rotated targets. Thirdly, a sparse representation model is introduced and the identification problem is converted to a l₁-minimization problem. Finally, different aircraft types are predicted based on an evaluation index, which is called residual error. To validate the aircraft identification method, a recorded infrared aircraft dataset is implemented in an airfield. Experimental results show that the proposed method achieves 98.3% accuracy, and recovers the identity beyond 80% accuracy even when the test images are corrupted at 50%.