The invention discloses an SoC-FPGA-based self-reconstruction K-means cluster technology realization method. The method comprises the following steps: S1, reconstructing an SoC-FPGA heterogeneous platform model through which an ARM host end cooperates with an FPGA equipment end; S2, the ARM end constructing an OpenCL host program, creating a core and finishing memory distribution and mapping; S3, the host program scheduling a core program of the FPGA equipment end to transmit data to the FPGA equipment end; S4, a first OpenCL core program calculating an Euclidean distance in a parallel pipelined mode, and generating a distance matrix; S5, self-reconstructing a second OpenCL core program, and screening a minimum element of each row and recording a corresponding mass center; S6, self-reconstructing a third OpenCL core program to realize distance accumulation and quantity statistics work of all sample points in each mass center cluster; S7, the host program calculating new mass center data; and S8, the host program performing iteration judgment. The method provided by the invention improves the execution speed of a K-means cluster algorithm, obtains higher energy efficiency, and solves the problem of insufficient FPGA hardware resources through self-reconstruction of the core.