Pooling processing method and system applied to convolutional neural network

Abstract

The invention discloses a pooling treatment method. The method is applied to a pooling processing system of a convolutional neural network. The pooling processing system comprises a first storage device, a data region, a pooling core and a pooling controller. The method comprises the following steps of: obtaining a sample; the pooling controller reads k pieces of feature data from the first storage device in each reading period, the read k feature data are written into the data area, the k feature data are arranged according to the first direction, in one updating period, nk data arranged according to the first direction are written into the data area, and the duration of the updating period is n times of the duration of the reading period; and after the updating period is finished, the data in the data area is transmitted to the pooling kernel for pooling operation, and the data in the data area comprises nk pieces of data and last m rows of data arranged in the second direction in the previous updating period before the nk pieces of data. According to the technical scheme, due to data multiplexing, the storage read-write frequency is reduced, and the pooling processing efficiencyis improved.

Description

technical field [0001] The present application relates to the field of computer technology, in particular to a pooling processing method applied to a convolutional neural network, a pooling processing system, a chip system, and a computer-readable storage medium. Background technique [0002] In recent years, with the continuous deepening of deep learning research, Convolutional Neural Network (CNN) can achieve accurate image recognition and classification, and has been widely used in medical imaging, security, biometric recognition, robotics, driverless and other fields. Become one of the main directions of deep learning. [0003] CNN usually includes two steps of convolution and pooling (Pooling). Convolution is to use a two-dimensional convolution kernel to slide on the input image or two-dimensional data with a configurable step size until the entire image or two-dimensional data is traversed. . The size of the convolution kernel is relatively small. In each sliding pr...

AI Technical Summary

Problems solved by technology

[0006] In order to solve the problem of low CNN processing efficiency due to pooling lag in the prior art, the embodiment of the present application provides a pooling processing method, which can realize data multiplexing through the data area, thereby reducing the number of data reads, Improve the efficiency of pooling processing, thereby improving the processing efficiency of CNN

Images