Method for acquiring ultra-light classification network model

Abstract

The invention discloses a method for acquiring an ultra-light classification network model, and the method comprises the steps: carrying out the architecture search of a neural network based on a specific classification task, and guaranteeing the acquisition of a high-precision lightweight network b1; further cutting off unvaluable filters in the b1 network through a fusion layer cutting and network pruning method, and obtaining b3; then a TensorRT graph optimization method is fused, the reasoning speed is further accelerated, meanwhile, a quantification method is adopted, the model is further lightened, and therefore the final ultra-light, ultra-fast and high-precision classification network model is obtained. According to the lightweight classification convolutional neural network model obtained by the method, the balance of the size, efficiency, resources and precision of the classification model can be realized, and the requirement of quickly and accurately identifying the image is further met; according to the method, a good identification effect can be obtained, and time and space are optimized due to the fact that the network structure is simplified.

Description

technical field [0001] The invention relates to the technical field of neural network models, in particular to a method for obtaining a classification network model. Background technique [0002] When the deep neural network is applied to the field of image recognition, since the deep neural network involves a large number of hyperparameters and complex structures, this complexity makes it very difficult to analyze the deep structure theoretically. Most of the work involves adjusting parameters or stacking more layers to Obtain better accuracy. Therefore, although the deep neural network has high accuracy, the calculation time and training time are long, which cannot meet the requirements of fast and accurate image recognition in various fields. Contents of the invention [0003] The technical problem to be solved by the present invention is to provide an ultra-lightweight classification network model acquisition method to achieve a balance between the classification model...

AI Technical Summary

Problems solved by technology

[0003] The technical problem to be solved by the present invention is to provide an ultra-lightweight classification network model acquisition method to achieve a balance between the classification model size, efficiency, resources, and accuracy, and to solve the problem of deep networks relying on expensive hardware configurations and time-consuming calculations and training. The problem of low accuracy of the width network further meets the requirements of fast and accurate image recognition