The management granularity of the traditional NAND flash memory garbage collection algorithm is block, and the subsequent development is based on page management, which improves the accuracy of calculating the heat of data at the cost of greater memory consumption. Taking into account the deficiencies of existing NAND flash garbage collection algorithms, a garbage collection algorithm based on block update sequence number is proposed. The proposed algorithm re-uses a block based management granularity, defines the update sequence number of the block, and presents a new method for calculating the heat of data based on dynamic threshold segmentation processing in order to perform effective cold and hot separation. At the same time, the victim block selection strategy adopts a new cost function to make a balance between efficiency and wear leveling. The experimental results show that while reducing system memory consumption, the proposed algorithm can achieve better garbage collection efficiency and wear leveling effect than GR, CB, CAT, FaGC, LRGC and LRGC+ algorithms.