CN-122023185-A - Enhancement processing method for definition of airway endoscope image

CN122023185ACN 122023185 ACN122023185 ACN 122023185ACN-122023185-A

Abstract

The invention provides an enhancement processing method of image definition of a respiratory tract endoscope, which belongs to the field of image processing and comprises the steps of downsampling a respiratory tract endoscope image, inputting the downsampled respiratory tract endoscope image into a constructed multi-scale feature extraction module for feature extraction and fusion, inputting the downsampled respiratory tract endoscope image into a constructed frequency enhancement fuzzy perception module for enhancement, finally performing transposition convolution operation and upsampling step by step, fusing the feature information of a shallow layer and a deep layer of a network to obtain the respiratory tract endoscope image enhancement image.

Inventors

WANG HUANHUAN
CUI CHUNGUANG

Assignees

中国人民解放军总医院第一医学中心

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (8)

1. The enhancement processing method for the definition of the airway endoscope image is characterized by comprising the following steps of: S1, collecting an image of an airway endoscope, and manufacturing an image dataset of the airway endoscope; s2, respectively performing 2 times and 4 times downsampling on the respiratory tract endoscopic image to obtain respiratory tract endoscopic image with 2 times downsampling and 4 times downsampling scales; S3, constructing a multi-scale feature extraction module FSDO, which comprises the steps of calculating the difference between any pixel in an input image and other pixels in a neighborhood, carrying out weighted integration by combining fractional distance weights and space random disturbance factors, carrying out normalization processing to obtain a feature extraction operator, and generating an output feature map by fusing the feature extraction operators of a plurality of scales through nonlinear weighting; S4, processing the image of the respiratory tract endoscope by FSDO, and obtaining a respiratory tract endoscope image characteristic diagram by upsampling and fusing the image of the respiratory tract endoscope image of the lower-layer scale with the image of the respiratory tract endoscope of the upper-layer scale after processing the image of the respiratory tract endoscope of the lower-layer scale; S5, constructing a frequency enhancement fuzzy perception module AFFRHO, which comprises the steps of carrying out two-dimensional fractional Fourier transform on an input feature map to obtain a frequency spectrum feature map, respectively enhancing the frequency spectrum feature map by adopting a random harmonic enhancement and frequency enhancement weighting mode, then fusing the frequency spectrum feature map, and converting the frequency spectrum feature map into an enhancement feature map through inverse fractional Fourier transform; S6, processing the image feature map of the respiratory tract endoscope by AFFRHO, then carrying out step-by-step up sampling by transposition convolution operation, and combining feature information of the shallow layer and the deep layer of the jump connection fusion network to obtain an enhanced image of the respiratory tract endoscope.
2. The method for enhancing sharpness of an endoscopic image according to claim 1, wherein in S3, the multi-scale feature extraction module FSDO is constructed, and the specific process includes: s31, for any pixel in the input image In its neighborhood And (3) calculating the difference between the pixel and other pixels, carrying out weighted integration by combining fractional distance weight and space random disturbance factor, and carrying out normalization processing to obtain a feature extraction operator, wherein the specific formula is as follows: , Wherein the method comprises the steps of As a differential term, Is a fractional order decay term of the distance, As a parameter of the scale of the fractional order, Is a small constant value, and is a constant value, As a function of the spatial randomness, For the normalized coefficient, the specific calculation mode is as follows: ; S32, generating an output feature map by fusing feature extraction operators of all scales through nonlinear weighting, wherein the specific formula is as follows: , Wherein the method comprises the steps of In order to fuse the weights, the weights are, As an arctan () function, As a function of the tanh, Is a spatially random signal.
3. The method for enhancing the definition of an endoscopic image according to claim 2, wherein in the step S4, the multi-scale feature extraction module FSDO is used to process three endoscopic image images of different scales respectively, and the feature fusion is performed on the lower-scale endoscopic image feature map and the upper-scale endoscopic image feature map after the up-sampling operation, so as to finally obtain three endoscopic image feature maps of different scales.
4. The method for enhancing sharpness of an endoscopic image of a respiratory tract according to claim 3, wherein in S5, the process of performing a two-dimensional fractional fourier transform on an input feature map to obtain the spectral feature map comprises: s51, for an input feature map, firstly constructing a kernel function of two-dimensional fractional Fourier transform And Wherein And The rotation parameters of fractional Fourier transform in the x and y directions are respectively calculated, and then two-dimensional fractional Fourier transform is carried out, wherein the specific formula is as follows: , Wherein the method comprises the steps of And the spectrum characteristic diagram is obtained after two-dimensional fractional Fourier transform.
5. The method for enhancing sharpness of an endoscopic image according to claim 4, wherein in S5, the process of using the random harmonic enhancement spectrum feature map is as follows: The random harmonic disturbance mechanism is introduced to enhance the frequency spectrum characteristics, and the detail change of the fuzzy area of the image is simulated, wherein the specific formula is as follows: , Wherein the method comprises the steps of As a random harmonic disturbance factor, Is a local spectral frequency estimate.
6. The method for enhancing sharpness of an endoscopic image according to claim 5, wherein in S5, the step of enhancing the spectral feature map by using the frequency enhancement weighting method is as follows: The frequency enhancement weighting strategy is adopted, weaker enhancement is adopted for a low-frequency region, stronger enhancement is adopted for a high-frequency region, and the enhancement weight is self-adaptively adjusted according to the magnitude of the frequency spectrum by calculating the magnitude of the frequency spectrum, wherein the specific formula is as follows: , Wherein the method comprises the steps of And In order to adjust the parameters of the degree of frequency enhancement, Is the frequency domain amplitude.
7. The method for enhancing the sharpness of an endoscopic image of a respiratory tract according to claim 6, wherein in S5, the spectral feature images are respectively enhanced by a method of random harmonic enhancement and frequency enhancement weighting and then fused, and the specific process of converting the spectral feature images into the enhanced feature images through inverse fractional fourier transform is as follows: The spectrum characteristics after random harmonic disturbance and frequency enhancement weighting are fused in a way of multiplying element by element to obtain an enhanced spectrum characteristic diagram, and then the enhanced spectrum characteristic diagram is converted back to a space domain through inverse fractional Fourier transform to generate an enhanced characteristic diagram, wherein the specific formula is as follows: , Wherein the method comprises the steps of And As a kernel function of the inverse fractional fourier transform, And Is a rotation angle parameter of the inverse fractional fourier transform.
8. The method for enhancing sharpness of an image of an airway endoscope according to claim 7, wherein in the step S6, the frequency enhancement blur perception module AFFRHO is used to process three image feature maps of the airway endoscope with different scales respectively to obtain three image enhancement feature maps of the airway endoscope with different scales, then the image enhancement feature maps of the airway endoscope are up-sampled step by using a transpose convolution operation, and shallow features and deep features are fused by jump connection, and finally the image enhancement image of the airway endoscope is obtained The output is decoded for the highest scale.

Description

Enhancement processing method for definition of airway endoscope image Technical Field The invention belongs to the field of image processing, and particularly relates to an enhancement processing method for the definition of an endoscopic image of a respiratory tract. Background The respiratory endoscope is widely applied to diagnosis and treatment of respiratory diseases, the quality of the image directly influences the diagnosis and treatment precision, however, in practical clinical application, due to uneven illumination of imaging equipment, scattering of biological tissues and noise of the equipment, the obtained image often has serious noise interference and fuzzy distortion, so that the observation and judgment of doctors on pathological details are seriously influenced, and the diagnosis and treatment accuracy is reduced. The traditional method for enhancing the image of the respiratory tract endoscope generally adopts a convolutional neural network, classical Fourier transform or wavelet analysis technology, and has the defects that a fixed local receptive field exists in convolutional operation, texture details are easy to lose due to the fact that the convolutional operation is smooth, wavelet analysis can only capture specific frequency components and cannot effectively adapt to complex multi-scale random textures in images, traditional Fourier transform only processes integer-order frequency spectrums and cannot finely sense harmonic attenuation characteristics caused by fuzzy, a contrast file with a publication number of CN118134950A discloses a semi-supervised bronchial image segmentation method and device based on double disturbance consistency, phase information related to high-frequency structural characteristics is reserved through FFT, amplitude information related to low-frequency semantics is changed to serve as a strong data enhancement mode, the relevance among multi-scale frequency domain characteristics is not considered, the structural information is easy to lose, a contrast file with a publication number of CN120526063B discloses a three-dimensional bronchial image generation device, global characteristics are calculated through multi-level self-attention, local characteristics are output through convolution calculation and wavelet transform, finally feature fusion is carried out, but the method is high in input image quality, the condition that the actual respiratory tract endoscope is not considered, the image is not required to be influenced, the condition of high-scale image quality is required to be improved, the image is difficult to be sensed, the condition is high, the image is difficult to be influenced by the local image is high, and the image is required to be influenced by the local image contrast, and has high contrast image quality. Disclosure of Invention Aiming at the defects of the prior art, the invention provides a method for enhancing the definition of an image of a respiratory tract endoscope, which aims at combining a multi-scale feature extraction module and a frequency enhancement fuzzy perception module to realize denoising enhancement of the image of the respiratory tract endoscope, and comprises the following steps: S1, collecting an image of an airway endoscope, and manufacturing an image dataset of the airway endoscope; s2, respectively performing 2 times and 4 times downsampling on the respiratory tract endoscopic image to obtain respiratory tract endoscopic image with 2 times downsampling and 4 times downsampling scales; S3, constructing a multi-scale feature extraction module FSDO, which comprises the steps of calculating the difference between any pixel in an input image and other pixels in a neighborhood, carrying out weighted integration by combining fractional distance weights and space random disturbance factors, carrying out normalization processing to obtain a feature extraction operator, and generating an output feature map by fusing the feature extraction operators of a plurality of scales through nonlinear weighting; S4, processing the image of the respiratory tract endoscope by FSDO, and obtaining a respiratory tract endoscope image characteristic diagram by upsampling and fusing the image of the respiratory tract endoscope image of the lower-layer scale with the image of the respiratory tract endoscope of the upper-layer scale after processing the image of the respiratory tract endoscope of the lower-layer scale; S5, constructing a frequency enhancement fuzzy perception module AFFRHO, which comprises the steps of carrying out two-dimensional fractional Fourier transform on an input feature map to obtain a frequency spectrum feature map, respectively enhancing the frequency spectrum feature map by adopting a random harmonic enhancement and frequency enhancement weighting mode, then fusing the frequency spectrum feature map, and converting the frequency spectrum feature map into an enhancement feature map through inverse fractional Four