To monitor the working state of a space target， attitude direction estimation of parabolic antenna loads from a multi-view sequence of the terahertz （THz） inverse synthetic aperture radar （ISAR） images is developed. A space-based THz radar imaging system， which aims to achieve surveillance of high earth orbit satellite targets and small satellite targets， is proposed. Under the theorem that the projection of the parabolic antenna edge （a circle） along arbitrary observation direction is an ellipse， an improved Randomized Hough Transform is proposed to automatically detect and calculate the five key parameters of ellipse components from each THz ISAR image. To ensure the efficiency， accuracy， and robustness of the estimated attitude direction， a two-level estimation algorithm is proposed. The radius and three-dimensional center location of the antenna edge are estimated first. Then， taking these parameters as prior information， the attitude direction is estimated by solving an optimization to minimize the joint error about the length of semi-minor axis and the inclination angle of an ellipse. Electromagnetic scattering data of satellite model targets illustrate the effectiveness and robustness of the proposed method in attitude direction estimation of parabolic antenna loads.