With the development and popularization of remote sensing imaging technology, a large number of bands in hyperspectral images make most application researches encounter Hughes phenomenon. With the rapid growth of hyperspectral image data, the computational complexity of the existing traditional serial algorithm is high, and it is difficult to deal with high-dimensional and massive hyperspectral image data. Aiming at the above problems, a band selection algorithm based on mutual information calculation under Spark platform is proposed. The band correlation and multiple correlation are defined by entropy and mutual information theory. The data column transformation is designed based on Spark RDD programming model, and the data set is divided into column matrix to reduce the computational load. The algorithm is parallelized on the Spark platform to improve algorithm execution efficiency. Experimental results show that the proposed algorithm achieves an overall classification accuracy of 94.5%±0.5, with good acceleration performance and improved data scalability.