Automatic search method for optimal structure of convolutional neural network

Abstract

The invention discloses an automatic search method for an optimal structure of a convolutional neural network, and belongs to the field of machine learning visual recognition. According to the method, for each specific computer vision task, an optimal convolutional neural network structure is automatically searched based on a random drift particle swarm algorithm; in the searching process, each particle position is set to represent a CNN structure, and through continuous iteration of the random drift particle swarm algorithm, a deep convolutional neural network with a smaller model and better performance is finally searched. The average classification accuracy of 98.16% is obtained when an image classification experiment is carried out on a Rectangles data set, and compared with the accuracy of 87.66% obtained by an existing LeNet-5 network of a 7-layer CNN structure and the classification accuracy of 88.96% obtained by an existing AlexNet network of a 11-layer structure, the CNN structure searched by the method can obtain higher accuracy.

Description

technical field [0001] The invention relates to an automatic search method for an optimal structure of a convolutional neural network, which belongs to the field of machine learning visual recognition. Background technique [0002] Deep convolutional neural network is a special type of neural network that can achieve almost better results than traditional machine learning algorithms in solving computer vision problems such as image classification, recognition, object detection or semantic segmentation. [0003] In recent years, in order to use deep learning methods to solve computer vision problems, classic Convolutional Neural Networks (CNN) models such as LeNet, AlexNet, VGG and ResNet have been proposed successively. However, when designing a CNN structure to solve a practical problem, you first need to determine the following parameters: what kind of CNN network structure to use, how many layers CNN uses, how many neurons are needed for each layer, etc.; that is, CNN nee...

AI Technical Summary

Problems solved by technology

[0005] In order to solve the time-consuming and labor-intensive problem of finding the optimal CNN structure for any specific computer vision task, the present invention provides an automatic search for the optimal structure of the convolutional neural network based on the random drift particle swarm optimization algorithm. method, said method comprising:

Images