Image classification processing method of neural network based on frequency domain wavelet basis processing

Abstract

According to the image classification processing method of the neural network based on frequency domain wavelet basis processing, frequency domain analysis is combined into image classification processing, operation is simple, classification precision is high, accuracy is good, and the method is closer to image processing of human vision. The method comprises the following steps: step 1, decomposing information in a natural image into three groups according to frequency band distribution by utilizing multistage discrete wavelet transform; and step 2, gradually injecting the obtained three groups of information into the neural network from low frequency to high frequency. An input image is decomposed through given wavelet transform, and coefficients of wavelet transform are gradually fed into different depth map layers of a deep neural network according to decomposition levels. Furthermore, while feed-in is carried out, an attention module is used for promoting fusion of neural networkfeatures and injection information, so that remarkable performance gain is generated. Therefore, how the information of different frequencies affects the accuracy of the deep neural network is obtained.

Description

technical field [0001] The invention relates to a neural network image processing method, in particular to a neural network image classification processing method based on frequency domain wavelet base processing. Background technique [0002] It is well known in the field of image processing that information can be decomposed into different frequencies, and each frequency has its advantages. However, existing neural networks always ignore these distinctions and directly feed all information into the neural network together, treating them equally. [0003] Deep convolutional neural networks (CNNs) have been widely used in various applications, such as image classification, object detection, and image segmentation. Despite impressive numerical achievements, roadblocks to efficient and accurate human-like vision systems remain. [0004] First, CNNs have a "black box" nature that makes it difficult to explain their predictions. Although great efforts have been made to explai...

AI Technical Summary

Problems solved by technology

[0004] First, CNNs have a "black box" nature that makes it difficult to explain their predictions

Images