Aiming at the problem of poor real-time performance of Brillouin optical time domain analysis (BOTDA) distributed optical fiber sensing system, we propose a extraction method of Brillouin scattering spectrum feature based on the cross-correlation convolution and high-order centroid calculation to reduce measuring time. Firstly, the sweep data of the Brillouin scattering spectrum along the optical fiber is convoluted with the ideal Lorentz curve, and the convolution result has a nearly ideal Lorentzian distribution around its peak. Then we carry out high-order centroid extraction, and take the extracted result as an estimated value of Brillouin frequency shift (BFS). A 1.5 km Rayleigh BOTDA temperature sensing experiment is designed to verify the feasibility of the proposed algorithm. The results show that the proposed algorithm is different from the conventional Lorentz curve fitting (LCF), it avoids extension of measuring time caused by the complex iterative process, and has good real-time performance and measurement accuracy. The proper selection of data point numbers and calculation orders can control the error within 1 MHz. When the sweep interval has to be increased to reduce the measurement time in long-range and high-resolution dynamic measurement, the proposed algorithm has a smaller measurement error than the LCF based on the Levenberg-Marquardt (LM) algorithm.