Construction method of weighted residual neural network

Abstract

The invention discloses a construction method of a weighted residual neural network, which comprises the following steps: sequentially connecting a convolution layer, a batch normalization layer and an activation layer from front to back, and conducting packaging to form a root module; packaging a main branch and a bypass branch which are parallel to each other to obtain a weighted incomplete module; the main branch is formed by packaging after smoothly repeating a convolution layer, a batch normalization layer and an activation layer for multiple times from front to back; if the weighted residual neural network performs down-sampling, the bypass branch is composed of a convolutional layer and a bypass branch network in parallel, wherein the bypass branch network is formed by packaging after smoothly repeating a convolution layer, a batch normalization layer and an activation layer from front to back, otherwise, the bypass branch is formed by packaging after smoothly repeating a convolution layer, a batch normalization layer and an activation layer for multiple times from front to back; a global average pooling layer, a full connection layer and an activation layer are sequentiallyconnected from front to back, and packaging is conducted to form a head module; and the root module, the plurality of weighted incomplete modules and the head module are sequentially connected to obtain the weighted residual neural network.

Description

technical field [0001] The invention relates to the field of image classification in computer machine vision, in particular to a method for constructing a weighted residual neural network. Background technique [0002] At present, neural network technology in computer vision is widely used in many fields such as image classification, target detection, image segmentation, face recognition, and behavior recognition. In these fields, image classification is the most basic technology. Most of the neural networks used in other fields use the neural network for image classification as their backbone network, which is implemented after adding other functional modules. Therefore, a high-performance image classification network is very important for machine vision based on neural network technology. [0003] Residual network, as a popular type of network in image classification network, mainly utilizes the residual module that can be stacked repeatedly. In the residual module, a b...

AI Technical Summary

Problems solved by technology

However, the information output by the main branch has passed through more network layers, and theoretically has richer feature expression information. It is directly added to the output of the bypass branch with fewer network layers, which may weaken the features extracted by the main branch. , thus affecting the improvement of model accuracy

Images