The Brillouin Optical Time-domain Analysis technology(BOTDA) is one of the hotspots in the optical fiber sensing field, it can measure temperature and strain information along the fiber based on the linear relationship between temperature, strain and Brillouin frequency shift. This technology has high measurement accuracy and long sensing distance. However, the Brillouin scattering signal in BOTDA is very weak and easily affected by external factors. In order to further improve the measurement accuracy and real-time performance of distributed optical fiber sensing system based on Brillouin scattering, this paper proposes a Brillouin scattering spectrum feature extraction scheme based on sparse constrainted cross-correlation iterative algorithm model. Different from the traditional cross-correlation convolution algorithm, this method changes the frequency distribution of the reference Lorenz signal by constructing a constraint model to improve the extraction precision of the Brillouin gain spectrum. In this paper, a 24.4km BOTDA temperature sensing experiment system is built. The experimental results show that compared with the conventional cross-correlation convolution algorithm and Levenberg -Marquardt Lorenz curve fitting algorithm, the accuracy of the Brillouin frequency shift extraction of this method is increased by about 5MHz, and the error of Brillouin frequency shift extraction can be controlled at 1MHz.Besides, the computational complexity of this method is far less than theLevenberg -Marquardt Lorenz curve fitting algorithm. Therefore, the sparse constrainted cross-correlation iterative algorithm proposed in this paper can effectively improve the measurement accuracy and real-time performance of the Brillouin optical sensor system.