Deep learning reasoning framework based on satellite-borne CPU and FPGA heterogeneous platform and implementation method

The invention discloses a deep learning reasoning framework based on a satellite-borne CPU (Central Processing Unit) and FPGA (Field Programmable Gate Array) heterogeneous platform and an implementation method. According to the invention, a logic arithmetic unit is realized by using a single processing unit mode, a convolution operation unit, a pooling unit, an up-sampling unit and a channel associated unit are combined and multiplexed, and a binary instruction is used for controlling data to be transmitted among the units, so that reasoning acceleration of a complex neural network model is realized, hardware resource consumption is reduced, and the reasoning speed of the neural network model is improved. And the model adaptability of the reasoning framework is also improved. In a network model coding format, the coding length of a single binary instruction is 1024 bits, the first 510 bits are valid bits, and the rest bits are reserved extension bits, and by adding a generation and analysis method of an exten