Image feature extraction method, device, apparatus, and readable storage medium

Abstract

The invention discloses an image feature extraction method. The method comprises the following steps: obtaining an original image; compressing each pixel value of the original image by using a compression value corresponding to a fixed-point number of bits to obtain a target image; obtaining a target image by using the compression value corresponding to the fixed-point number of bits. The target image is input into a fixed-point depth convolution neural network model. The depth convolution neural network model is used to compute the fixed point of the target image, and the image features of the target image are obtained. Fixed-point computation is faster than floating-point computation in computational speed, and it is easier to be realized in practical application. In feature extraction,fixed-point computing can also reduce the storage overhead, reduce the occupation of computer resources, further improve the computing speed, and then can be used for real-time image feature extraction. The invention also discloses an image feature extraction device, apparatus and a readable storage medium, which have corresponding technical effects.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to an image feature extraction method, device, equipment and readable storage medium. Background technique [0002] The image feature extraction application designed by the algorithm dominated by deep neural network is widely used in people's daily work and study. [0003] A deep neural network is often composed of dozens or even hundreds of convolutional layers, and the feature maps generated during the calculation process need to occupy a large amount of storage space. In addition, the weight floating-point calculation cannot realize the processing effect of real-time image feature extraction. All of these lead to slow calculation speed of image feature extraction. [0004] However, the current compression of deep convolutional neural network models is mainly through fixed-point quantization of feature maps, especially the method of compressing models through re...

AI Technical Summary

Problems solved by technology

In addition, the weight floating-point calculation cannot realize the processing effect of real-time image feature extraction

This all leads to slow calculation speed of image feature extraction

[0004] However, the current compression of deep convolutional neural network models is mainly through fixed-point quantization of feature maps, especially the method of compressing models through regular or irregular quantization methods between layers to improve the speed of image feature extraction. The processing accuracy of the neural network model cannot be guaranteed, so that the accuracy of the extracted image features cannot meet the needs of users

Images