Image classification neural network model building and image classification methods and devices

Abstract

Embodiments of the invention provide image classification neural network model building and image classification methods and devices. The image classification neural network model building method comprises the steps of firstly, acquiring a training set containing classified images; according to the distance between convolutional layers in dense blocks of a densely connected convolutional neural network model, determining the number of characteristic diagrams of the convolutional layers in the input dense blocks; forming a second densely connected convolutional neural network model; training the second densely connected convolutional neural network model by using the training set to generate an image classification neural network model; and classifying to-be-classified image data by using the image classification neural network model. According to the image classification neural network model building and image classification methods and devices, network parameters in an image classification process are reduced.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method and device for establishing an image classification neural network model and image classification. Background technique [0002] Image classification technology has a wide range of applications in character recognition, face recognition, object recognition, pedestrian detection, image retrieval, etc. The main idea of ​​image classification is: first analyze the target or scene contained in the sample image, then convert the original pixels of the image into feature information of the image, use these feature information to represent the description information of each image, and finally according to the obtained These descriptive information are used for image classification. The traditional method of image classification has limitations. It needs to extract and classify image features separately, and the extraction of image features is the premise of accurate ...

AI Technical Summary

Problems solved by technology

[0004] In view of the above analysis, the embodiment of the present invention proposes an image classification neural network model and an image classification method and device to solve the existing densely connected convolutional neural network. With the widening of the network width and deepening of the depth, The problem of explosive growth in the number of network parameters

Images