Data processing method and device for low computing power processing equipment

Abstract

The invention discloses a data processing method and device for low computing power processing equipment. The method comprises the following steps: in a real-time computer vision processing process, processing equipment with low computing power acquires image data; the processing equipment performs computer vision processing on the acquired image data by using a preset neural network to obtain a computer vision processing result, wherein the preset neural network is a target neural network obtained through the following processing. An initial neural network is constructed, a plurality of preset specific structures in the initial neural network are provided with corresponding sparse scaling operators respectively, and the sparse scaling operators are used for scaling output of the corresponding specific structures; the weight of the initial neural network and a sparse scaling operator of a specific structure are trained by adopting preset training sample data to obtain an intermediate neural network; and a specific structure of which the sparse scaling operator is zero in the intermediate neural network is deleted to obtain a target neural network.

Description

technical field [0001] The present invention relates to the field of computers, in particular to a data processing method and device for processing equipment with low computing power. Background technique [0002] In recent years, deep neural networks have achieved great success in many fields, such as computer vision, natural language processing, etc. However, the model of deep neural network often contains a large number of model parameters, and the calculation amount is large and the processing speed is slow. A device with low computing power means that if the computing power of a device is lower than the computing power required by the computing task or computing model deployed on it, the device is a device with low computing power. [0003] To solve this problem, some solutions are currently proposed: [0004] Scheme 1. Hao Zhou pointed out in the paper "Less is More: Towards Compact CNNS" that the existing convolutional neural network includes a large number of param...

AI Technical Summary

Problems solved by technology

[0007] However, the neural network obtained according to the above scheme still cannot better meet the requirements of compact structure, fast running speed, and high precision, so it cannot perform real-time calculations on devices with low computing power.

Images