High-speed pooling method based on FPGA

The invention discloses a high-speed pooling method based on an FPGA, belongs to the technical field of deep learning intelligent algorithms, and solves the technical problems that in the prior art, pooling operation consumes on-chip storage resources, and real-time performance cannot be guaranteed. The method comprises the following steps: segmenting an image to be processed according to rows, and sequentially storing segmented sub-images according to a row sequence; starting from the first row, taking every M rows of sub-images as a group of to-be-processed sub-images; the size of the pooling core is M * M; receiving and pooling each group of to-be-processed sub-images; when each group of to-be-processed sub-images is received, synchronously inputting M rows of sub-images in the group of to-be-processed sub-images into a block-shaped storage unit with the size of M * M in the FPGA chip according to columns; when the block-shaped storage unit is full of M * M data, pooling processing is carried out, and a po