Automatic grouping multi-scale lightweight deep convolutional neural network optimization method

Abstract

The invention discloses an automatic grouping multi-scale lightweight deep convolutional neural network optimization method. The method comprises the steps: 1, performing evolution search by adopting a symmetric sine and cosine algorithm to obtain a to-be-evaluated population; 2, performing fitness evaluation on the population to be evaluated on an image classification task by using a multi-scale lightweight convolutional neural network to obtain an evaluated population; 3, decoding the best individual in the evaluated population to obtain a final network model SCA_Mblock Net, and performing model performance evaluation. The automatic grouping technology provided by the invention is applied to a feature fusion module of a multi-scale convolution structure of an original deep convolutional neural network, an improved multi-scale lightweight deep convolutional neural network is designed, and the multi-scale lightweight deep convolutional neural network is obtained without domain knowledge and manual intervention. The average accuracy of an original convolutional neural network on an image classification data set can be improved by 2.56%, and compared with several advanced competitors of the same kind, the method has remarkable advantages.

Description

technical field [0001] The invention belongs to the research field of evolutionary deep convolutional neural network structure optimization technology, and specifically relates to an automatic grouping multi-scale lightweight deep convolutional neural network optimization method. Background technique [0002] Deep convolutional neural network (DCNN) is a kind of deep neural network (DNN), which has been widely used and developed in the fields of computer vision (CV) and image processing because of its excellent feature extraction ability. Convolutional neural network (CNN) is one of the most suitable learning algorithms for image content representation, and has shown excellent performance in image segmentation, object detection, and image classification tasks. For example, in the image classification task, from the advent of LeNet in 1995, to the outstanding performance of AlexNet in the ImageNet Challenge in 2012, to the proposal of VGGNet in 2015, ResNet and DenseNet in 20...

AI Technical Summary

Problems solved by technology

The performance of the convolutional neural network optimized by using the genetic algorithm has been improved to a certain extent, but the design of the crossover process is still difficult due to the variable-length coding strategy, and the crossover process itself is designed for fixed-length coding. Therefore, the progeny after crossover by variable length coding may have the limitation of insufficient diversity

Images