Abstract: With the rapid development of commercial spaceflight, the demand for rocket attitude estimation based on external measurement data has become increasingly urgent. In this paper, an in-depth analysis of the characteristics of high-resolution range profile (HRRP) of rocket is first conducted. To address the issues of high noise and fluctuation in HRRP data, the HRRP of rocket is then reconstructed using a truncated mixture of Gaussians models based on a parameterized representation approach. Considering different scenarios of whether the rocket model is known, two methods: template matching and projection estimation are developed to achieve attitude estimation under both cases with and without a template library. Meanwhile, a neural network architecture is introduced combining multi-layer perceptrons and convolutional neural networks in different configurations for attitude estimation. Experimental results based on both simulated and flight HRRP sequences demonstrate that the reconstructed HRRP exhibits high signal-to-noise ratio and strong stability. By integrating multi-envelope detection, distance-based template matching, and length projection, the rocket separation detection and attitude estimation using HRRP data are successfully and stably performed. The study in this paper provides a valuable reference for future work on attitude estimation of rockets using external measurements.