Remote eye visual light diagnosis and treatment intelligent service method and system and storage medium

Abstract

The invention relates to the field of image processing, and discloses a remote eye visual light diagnosis and treatment intelligent service method and system and a storage medium, and the method comprises the steps: obtaining a to-be-processed eye fundus image, and carrying out the standardization processing of the image, and the standardization processing comprises the normalization processing and geometric correction of the gray value of the image; performing edge detection processing on the standardized image, extracting a structural feature region contained in the image, and dividing image sub-regions based on the structural feature region; the system comprises an image acquisition and processing module; an encoding and transmission module; a decoding and recovery module; and an evaluation and diagnosis module. According to the method, the region weight map and the multi-scale decomposition technology are introduced, so that the image reconstruction precision is remarkably improved, the detail retention capability is optimized, higher-quality image restoration is realized, and the method is particularly excellent in complicated images and scenes with rich details.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and specifically to a remote ophthalmology diagnosis and treatment intelligent service method, system and storage medium. Background Art

[0002] In modern society, image data processing and analysis play a vital role in various fields. For example, in applications such as medical imaging, video surveillance, and autonomous driving, accurately and quickly recovering image information while preserving image detail and structure is a core issue. Image reconstruction technology is particularly crucial in scenarios such as image compression, denoising, and missing data filling. While existing image encoding and decoding technologies can accomplish basic reconstruction tasks, as demand continues to increase, handling complex image details and preserving the integrity of important information remains an unresolved issue.

[0003] Variational autoencoders, a widely used image coding technique, offer advantages in image compression and generation. By mapping an image into a latent space, they effectively compress image information while preserving the image's overall structural characteristics to a certain extent. This advantage lies in efficient image reconstruction using a lower-dimensional latent representation, reducing computational resources. During the decoding phase, the variational autoencoder employs standard upsampling and deconvolution techniques to restore the image's general structure.

[0004] However, the existing technology still has some deficiencies, which limit its application in more complex image processing tasks; first, the traditional VAE model usually ignores the different importance of various regions in the image when encoding features, which makes the information in key areas easily lost or over-compressed. Secondly, although some models have tried to introduce multi-scale processing, these methods still do not effectively solve the problem of expressing image details and structures. In the decoding stage, existing technologies often rely on a single-scale reconstruction path, which makes the local details and high-frequency information of the image easily distorted, resulting in poor quality of the reconstructed image, especially in images with rich or diverse details. In addition, the training optimization of existing technologies usually focuses on pixel-level reconstruction loss, ignoring the adaptive optimization of the image at the structural level, resulting in the loss of structural information and blurring of details during image reconstruction. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a remote ophthalmology diagnosis and treatment intelligent service method, system and storage medium, which solves the problems of neglecting regional information and insufficient multi-scale feature expression in the image reconstruction process in the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: a remote ophthalmology diagnosis and treatment intelligent service method, comprising the following steps: Acquiring a fundus image to be processed and performing standardization processing on the image, wherein the standardization processing includes normalizing the grayscale value of the image and performing geometric correction; Performing edge detection processing on the standardized image to extract regions containing structural features in the image, and dividing the image into sub-regions based on the structural feature regions; Calculating information entropy values ​​for the image sub-regions respectively, and generating a region weight map based on the information entropy values ​​for weighted configuration of compression processing for different image sub-regions; Performing feature encoding on the standardized image based on a variational autoencoder, and using the region weight map to adjust the encoder's generation of a latent space representation of each image subregion to generate corresponding latent coding data; The latent coding data is used as input for decoding processing, and image features of the image at different scales are extracted through a multi-scale decomposition method. The image is then subjected to variational decoding and reconstruction based on the image features at each scale to generate an image reconstruction result.

[0007] Preferably, the normalization processing and geometric correction specifically include: Normalizing the grayscale value of the fundus image so that the pixel value range of the image is mapped to the specified grayscale range; Geometric correction involves using affine transformation and perspective transformation to correct the distortion in the image caused by shooting angle and deformation.

[0008] Preferably, the structural feature area specifically includes: The optic disc, macula, and blood vessel areas in fundus images contain high texture information; The extraction of the structural feature region adopts an edge detection algorithm to accurately extract key edge and contour information in the image.

[0009] Preferably, the information entropy value specifically includes: Calculate the entropy value of each image sub-region. The entropy value calculation formula is: ; in, is the entropy value of the image area; is the pixel value in the image area The probability distribution of is the number of pixel values; The first Pixel value; Regions with high entropy values ​​contain more information; In the process of calculating the information entropy value, the probability distribution of each region is calculated through the grayscale histogram.

[0010] Preferably, the regional weight map specifically includes: Generate a region weight map based on the entropy value of each image sub-region; The regional weight map is generated in the following way: ; in, For location The regional weight value at ; Image area The entropy value of is the maximum entropy value of all regions in the image; By using the region weight map, different compression ratios are applied to different regions in the subsequent compression step.

[0011] Preferably, the feature code specifically includes: Encoding the standardized image based on a variational autoencoder, wherein the encoding process includes extracting deep features of the image through a convolutional neural network and generating a latent space representation through a fully connected layer; During the feature encoding process, a variational inference method based on reparameterization techniques is used to optimize the distribution of the latent space.

[0012] Preferably, the potential coding data specifically includes: The latent space representation data generated in the encoding step represents low-dimensional features of the fundus image, and the feature encoding data is composed of multiple encoding vectors; The latent coding data is further quantized to convert it into discrete data of fixed size; The latent coded data is obtained by adopting the latent space of variational autoencoder To represent it, including the two parameters of mean and variance, it is used to reconstruct the probability distribution of the input image.

[0013] Preferably, the multi-scale decomposition method includes: Perform multi-scale wavelet transform on the input image, decompose the image into multiple frequency components, and encode, decode and reconstruct these frequency components respectively; The multi-scale decomposition method extracts information of the image at different scales; The multi-scale decomposition adopts discrete wavelet transform and processes the low-frequency part and the high-frequency part of the image respectively.

[0014] It also provides a remote ophthalmology diagnosis and treatment intelligent service system, including: Image acquisition and processing module, used to acquire fundus images and perform image standardization, identify key areas and perform weighted processing on the areas according to entropy values; The encoding and transmission module encodes the weighted fundus image based on the variational autoencoder and transmits the compressed data to the remote end; The decoding and recovery module is used to decode the compressed data received at the remote end and restore image details using multi-scale variational reconstruction technology; The evaluation and diagnosis module is used to evaluate the quality of the restored images and provide image data to remote doctors to support ophthalmic diagnosis.

[0015] A storage medium is also provided, which stores a computer program. When the computer program is executed by a processor, a remote ophthalmology diagnosis and treatment intelligent service method can be implemented.

[0016] The present invention provides a remote optometry diagnosis and treatment intelligent service method, system, and storage medium. It has the following beneficial effects: 1. This paper employs a method for generating latent image representations based on a variational autoencoder and introduces a region weight map to guide the encoder, achieving differentiated compression representation of structural regions in the image. Compared to traditional encoding schemes without differential processing, this method effectively avoids the loss of key structural information during the encoding process and addresses the problem of uneven representation of local image details.

[0017] 2. This invention introduces a multi-scale decomposition approach to extract image features during the decoding process, constructing a richer layered representation of information during structural restoration. Compared to existing techniques that rely solely on a single-scale decoding path, this method improves the ability to restore different layers of texture and contours in complex images, avoiding issues such as overall image blur or loss of detail.

[0018] 3. The combined strategy of region guidance and multi-scale reconstruction proposed in this paper enables image structural perception in both the encoding and decoding stages. Compared with traditional end-to-end black-box feature compression, it improves the structural constraint mechanism in the latent space, effectively enhancing the model's adaptability to semantic differences in image regions, and fundamentally addressing the technical shortcomings of coexistence of information redundancy and compression distortion.

[0019] 4. The present invention introduces a scale-aware reconstruction loss function in the decoding stage and improves the model's ability to fit the original image structure through hierarchical alignment in the training stage. This approach is different from the conventional image reconstruction mechanism that only relies on pixel-level MSE errors. It makes the model pay more attention to the consistency of the global structure and effectively overcomes the pain point of difficulty in recovering high-frequency information. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Schematic diagram of the method flow of the present invention; Figure 2 Schematic diagram of the system architecture of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] Please see the attached Figure 1 , an embodiment of the present invention provides a remote ophthalmology diagnosis and treatment intelligent service method, comprising the following steps: S1. Acquire a fundus image to be processed and perform standardization on the image, wherein the standardization includes normalizing the grayscale value of the image and performing geometric correction; In remote ophthalmology and optometry diagnosis and treatment intelligent service systems, image processing is a key preprocessing step, especially the standardization of fundus images after acquisition. This step aims to ensure that images from different sources have the same grayscale standard and geometric consistency, providing reliable input for subsequent feature extraction, encoding, compression, and diagnosis.

[0023] Image standardization involves two key steps: grayscale normalization and geometric correction. These two steps not only help eliminate inconsistencies caused by differences in equipment and environment during image acquisition, but also ensure the geometric accuracy of the images for precise analysis and diagnosis of fundus structures.

[0024] In this embodiment, after acquiring the fundus image to be processed, the image is first normalized, specifically by grayscale normalization and geometric correction. This normalization process helps eliminate image grayscale inconsistencies caused by external factors such as equipment and lighting, ensuring a uniform grayscale range for subsequent image processing. Geometric correction also corrects for image distortion caused by the image's shooting angle and equipment, ensuring accurate representation of fundus structures in the image.

[0025] In this embodiment, the grayscale values ​​of the images are first normalized. Since the grayscale value ranges of different images may be different, the normalization process can map the grayscale values ​​of the images to a standard grayscale range to eliminate the grayscale differences between images.

[0026] Generally, the grayscale value range of fundus images varies depending on factors such as the acquisition device, shooting angle, and ambient lighting. In order to eliminate these differences, this embodiment uses a linear normalization method to adjust the grayscale value of the image to a uniform range. Specifically, assuming that the grayscale value of the original image is , whose minimum and maximum values ​​are and , through linear transformation, the gray value of the image Convert to normalized value : ; in, is the normalized grayscale value. This method brings images from different sources into the same grayscale range, eliminating grayscale inconsistencies between images caused by device differences or inconsistent shooting conditions.

[0027] In some embodiments, the normalization process may further employ logarithmic transformation or histogram equalization to enhance the contrast of the image, thereby better highlighting the fundus detail information, especially in low-contrast areas.

[0028] Geometric correction of images is another important standardization step. The geometric structure of acquired fundus images may deviate due to differences in device shooting angles or distortion in the image acquisition device. To ensure geometric consistency of the images, this embodiment employs geometric correction technology.

[0029] Geometric correction can be achieved through a variety of transformations, common methods include affine transformation and perspective transformation. In one possible implementation, affine transformation is first used to correct the image's translation, rotation, and scaling. The transformation matrix of the affine transformation is: It can be expressed as: ; in, , and , are the coordinates of a point in the original image and the corrected image respectively; , , , are the parameters in the affine transformation matrix; , is the translation amount. This affine transformation can handle issues such as rotation and translation in the image and ensure that the image is aligned in space.

[0030] For more complex geometric distortion, this embodiment also uses perspective transformation. Perspective transformation is used to deal with image perspective distortion caused by incorrect shooting angle or device optical characteristics. The transformation matrix of perspective transformation is: as follows: ; in, , and , are the coordinates of a point in the original image and the corrected image respectively; - The coefficients of each element in the perspective transformation matrix represent the combination of transformation factors such as scaling, rotation, translation, and projection in the transformation.

[0031] Perspective transformation can correct the geometric distortion in the image, restore the actual viewing angle of the image, and make the fundus structure in the image more realistic.

[0032] Alternatively, geometric correction methods based on feature matching can be employed. Image alignment is achieved by extracting feature points from the image, such as corners and edges, and using these features for registration. Common feature extraction algorithms include SIFT (Scale-Invariant Feature Transform) and SURF (Speeded Up Robust Features). After feature matching, a least-squares optimization algorithm is used to calculate the image transformation matrix and apply it to the geometric correction.

[0033] S2. Perform edge detection on the standardized image to extract regions containing structural features in the image, and divide the image into sub-regions based on the structural feature regions; After completing fundus image normalization, edge detection is necessary to further locate key visual structural areas within the image. This process aims to extract regions containing structural features, such as the outlines of the optic disc, blood vessels, and macula, and then divide the image into several subregions based on these features for subsequent feature extraction, image encoding, or lesion identification.

[0034] Generally speaking, accurate identification of structural edges can help identify local areas of medical diagnostic significance in an image. In actual processing, edge detection not only locates structural areas but also provides basic spatial information support for the division of image sub-regions.

[0035] In this embodiment, after image normalization is completed, edge detection is immediately performed on the normalized image to identify the boundaries of structural features in the image. This edge detection can be performed based on the gradient strength of the image's grayscale changes. By calculating the grayscale changes of each pixel in the image in the horizontal and vertical directions, locations with sharp grayscale changes are identified as potential edge points.

[0036] Specifically, let the normalized image be ,in is the two-dimensional spatial coordinate of the image, and edge detection can be achieved by the following gradient calculation: ; ; in, Indicates that the image is The gradient value of the direction; Indicates the gradient value in the direction; is the gradient amplitude of the point, which is used to measure the edge strength of the position; For the image Partial derivatives in direction; For the image Partial derivatives in the direction.

[0037] In one possible implementation, edge detection can be performed using the Sobel operator, Prewitt operator, or Canny edge detection method. Alternatively, Canny edge detection is preferred in this embodiment due to its high edge location accuracy and noise immunity. The Canny algorithm includes four main steps: Gaussian filtering, gradient calculation, non-maximum suppression, and dual-threshold edge connection.

[0038] After edge detection is complete, the extracted edge regions are further processed to identify image regions with structural features. For example, based on closed edge contours, the spatial region encompassing the optic disc boundary, major blood vessel courses, and the macular dark area is extracted. To ensure the accuracy of the structural regions, edge connectivity can be assessed by combining a region growing algorithm or a contour tracking method based on edge morphology.

[0039] In some embodiments, in order to enhance the structural perception ability of sub-region division, an edge-guided watershed algorithm can be introduced, combined with the gradient map Construct a watershed marker map to achieve stable segmentation of structure boundaries.

[0040] As an extension, in the process of dividing the image into sub-regions, statistical characteristics such as pixel density, average grayscale, and edge complexity in the region can also be combined for optimization and judgment to ensure the spatial closure and anatomical rationality of the divided sub-regions.

[0041] Specifically, after completing the subregion division, each subregion can be labeled and numbered for subsequent region selection, local feature extraction, or image compression processing. Each subregion can also be accompanied by its edge morphology information, such as perimeter and boundary curvature, for structural complexity analysis.

[0042] S3. Calculate information entropy values ​​for each image sub-region, and generate a regional weight map based on the information entropy values ​​for weighted configuration of compression processing for different image sub-regions; After extracting image structural feature regions and dividing them into subregions, in order to further implement more targeted image compression strategies, it is necessary to evaluate the differences in the value of information expression in each subregion. Based on this, the present invention introduces an information entropy measurement mechanism to quantitatively reflect the complexity and detail richness of image content in different image subregions, and constructs a regional weight map based on the information entropy value. This regional weight map is then used to control the weighting configuration during the image compression process, thereby optimizing the strategy of retaining more image information in important areas and achieving higher compression ratios in less important areas.

[0043] Generally speaking, areas of an image that contain rich texture or structural details have a more dispersed pixel grayscale distribution and are relatively more important to the overall image representation. These areas tend to have higher information entropy values. Therefore, weighting regions based on information entropy values ​​is an effective method for configuring compression weights based on image statistical characteristics.

[0044] In this embodiment, after the image sub-regions are divided, the corresponding information entropy value is calculated for each image sub-region.

[0045] The regional weight map can be generated by: Calculate the entropy value of each image sub-region. The entropy value calculation formula is: ; in, is the entropy value of the image area; is the pixel value in the image area The probability distribution of is the number of pixel values; The first Pixel values.

[0046] As an option, if the image is a color image, the information entropy can be calculated for each channel separately, and then the comprehensive information entropy can be obtained by weighted average: ; in, 、 、 Represents the information entropy on the red, green and blue channels respectively; , , is the weight coefficient of each channel.

[0047] After the information entropy calculation of all sub-regions is completed, a regional weight map is further generated. This weight map is used to guide the compression ratio configuration adopted by each region during the compression process. Specifically, let the weight map be , which is defined as follows: ; in, For location The regional weight value at ; Image area The entropy value of is the maximum entropy value of all regions in the image.

[0048] As an implementation strategy, in the image compression module, for each sub-region, according to its corresponding weight value , dynamically adjust compression parameters. For example, in a compression algorithm based on transform coding (such as DCT or DWT), more high-frequency coefficients are retained to reduce the loss of structured areas, while a higher quantization step size is implemented for low-weight areas.

[0049] In some embodiments, the region weight map can be used to generate a corresponding compression parameter mapping table for direct use by subsequent image encoders. The mapping table contains the compression level identifier, quantization parameter, and level of retained accuracy for each region, ensuring that the compression algorithm can be flexibly configured based on the importance of the region.

[0050] S4. Feature encoding is performed on the standardized image based on a variational autoencoder, and the region weight map is used to adjust the encoder's method for generating a latent space representation of each image subregion to generate corresponding latent coding data. After extracting structural feature regions, partitioning subregions, and constructing an information entropy weight map for a standardized image, the present invention further introduces an image compression coding mechanism based on a variational autoencoder (VAE) to achieve adaptive image feature encoding. This mechanism learns the latent space representation of the image to generate a structure. While compressing and preserving key image feature information, it also adjusts the encoder's regional perception capabilities in conjunction with the generated regional weight map, achieving differentiated feature abstraction and latent coding optimization.

[0051] Traditional image coding methods generally struggle to fully exploit the semantic differences between image regions, leading to information loss or ineffective redundant encoding of structurally important regions. This paper introduces a variational autoencoder and combines it with a region weight map to control the encoder's perception of each subregion and the encoding granularity, making feature extraction and compression coding more content-sensitive and adaptable.

[0052] In this embodiment, based on the completed image normalization, structural edge extraction, subregion division, and region weight map construction, the image is input into a variational autoencoder model for feature encoding. The variational autoencoder mainly consists of two parts: an encoder network and a decoder network. The encoder maps the input image to a set of distribution parameters in a latent space, typically including a mean vector and a log-variance vector.

[0053] Specifically, the noise vector is sampled from a standard normal distribution via the reparameterization trick , and calculate the potential representation : ; in, and are the mean and standard deviation vectors of the encoder output respectively; is the sampling noise term.

[0054] In one possible implementation, in order to make the latent representation generated by the encoder more structure-sensitive in terms of spatial distribution, the present invention uses the aforementioned region weight map to regulate the encoder structure. When the image enters the encoder, the region weight map is introduced as an additional guiding input and fused with the image feature map. For example, in the feature extraction layer of the encoder, the weight map is introduced. Perform channel-by-channel or pixel-by-pixel weighted superposition processing with the feature map to improve the model's perception of high-weight areas.

[0055] Alternatively, during the encoder feature extraction phase, the weight map can be spliced ​​into the input image feature tensor in an extended channel manner, allowing the model to automatically learn the impact of the region weighting mechanism on the distribution of latent representations during encoding. This approach can enhance the model's spatial selectivity while maintaining structural stability.

[0056] In this embodiment, the potential representation output by the encoder is Designed as a multidimensional feature vector, it compresses important structural information representing the entire image. For different sub-regions within the image, the regional weights are reflected in the latent space as a strengthened distribution constraint on the dimensions of specific regions. This means that the latent encoding dimensions of high-weighted regions tend to be more concentrated, resulting in higher information density, thus demonstrating better restoration capabilities in subsequent reconstruction.

[0057] In some embodiments, to improve the model's accuracy in encoding latent features in local regions, a regional attention mechanism can be implemented in the encoder's intermediate layers to automatically enhance the network's expressiveness in key regions based on the image weight map. This attention mechanism dynamically adjusts the response values ​​of feature channels in the intermediate layers, making the model more sensitive to feature learning in high-entropy regions.

[0058] Furthermore, during the distribution fitting process in the latent space, the present invention maintains the core optimization objective of VAE, namely, maximizing the Evidence Lower Bound (ELBO). This optimization objective includes a reconstruction error term and a KL divergence term, which respectively measure the difference between the reconstructed image and the input image, and the difference between the latent distribution and the standard normal distribution. The introduction of a weight map further adjusts the region-aware function in this optimization objective, thereby enabling backpropagation of the reconstruction error gradient based on region importance.

[0059] S5. Decoding the latent coded data as input, extracting image features at different scales through multi-scale decomposition, and performing variational decoding and reconstruction on the image based on the image features at each scale to generate an image reconstruction result. After completing the feature encoding of the standardized image and generating a latent space representation based on the regional weight map, the latent coded data needs to be decoded and reconstructed to further restore the image information and meet the input requirements of subsequent visual analysis or diagnostic applications. Based on this, the present invention introduces a multi-scale decomposition and fusion mechanism. By extracting image features at multiple scales during the decoding phase and combining the features at each scale for hierarchical reconstruction, the structural expression capability and detail preservation characteristics of the reconstructed image are enhanced. This processing strategy not only achieves efficient connection with the aforementioned encoding module, but also provides a structural perception basis for reconstructed image quality control.

[0060] Generally speaking, images exhibit different types of information at different scales. Low-scale layers primarily contain global structural features, while high-scale layers emphasize edges, textures, and local details. Therefore, introducing a multi-scale processing mechanism during image decoding helps to more comprehensively recover the information content carried by the latent representation in both the spatial and frequency domains.

[0061] In this embodiment, the latent coding data generated in the previous stage is used as the decoding input, and the latent representation is re-parameterized to obtain the latent vector for decoding. The latent vector first undergoes a multi-layer upsampling operation (e.g., deconvolution or pixel permutation module) in the decoder network for preliminary image restoration. To improve decoding quality, a multi-scale decomposition module is introduced into the decoder structure to extract feature maps layer by layer from the intermediate features during the decoding process, forming image features at multiple scale levels.

[0062] Alternatively, based on the scale extraction described above, a feature fusion module can be used to perform cascade, weighted, or attention fusion processing on image features at each scale. This process can use channel attention or spatial attention mechanisms to achieve adaptive weight learning between scales, thereby enhancing the reconstruction expressiveness of key areas.

[0063] Specifically, in the fusion operation, the feature map of each scale can be set Corresponding to a learnable fusion weight , the final image feature representation It can be expressed as: ; in, The total number of feature layers in the network involved in loss calculation.

[0064] In some embodiments, in order to maintain structural coherence, an upsampling interpolation operation may be introduced before the fused features are fed into the last layer of decoding module to keep the size of feature maps at all scales consistent, thereby facilitating subsequent convolution to restore the final image size.

[0065] In this embodiment, the fused image features are then processed by the decoding tail module to generate an image reconstruction result. This reconstruction result has the same spatial size as the normalized image and is optimized through a reconstruction error control mechanism during training (such as the MSE loss, SSIM loss, or perceptual loss) to achieve the desired balance between structural and semantic restoration.

[0066] As an implementation strategy, the loss function design within the variational decoder architecture introduces a scale-aware error term based on the reconstruction error. This term measures whether image features at each scale are effectively restored during the reconstruction process. This is accomplished by performing the same-scale decomposition on the original image and the reconstructed image, and calculating the error between the corresponding layers.

[0067] For example, suppose the first The layer features are , the corresponding layer feature of the reconstructed image is , then the scale-aware loss is: ; in, is the scale-aware loss function; The total number of feature layers in the network involved in loss calculation; For the The weighting coefficient of the layer.

[0068] This loss term is combined with the traditional image reconstruction loss function to form the final optimization objective function, ensuring the effective alignment and restoration of image features of different scales during the decoding process.

[0069] The remote ophthalmology diagnosis and treatment intelligent service system described below and the remote ophthalmology diagnosis and treatment intelligent service method described above can be referenced to each other.

[0070] Please see the attached Figure 2The present invention also provides a remote ophthalmology diagnosis and treatment intelligent service system, including: Image acquisition and processing module, used to acquire fundus images and perform image standardization, identify key areas and perform weighted processing on the areas according to entropy values; The encoding and transmission module encodes the weighted fundus image based on the variational autoencoder and transmits the compressed data to the remote end; The decoding and recovery module is used to decode the compressed data received at the remote end and restore image details using multi-scale variational reconstruction technology; The evaluation and diagnosis module is used to evaluate the quality of the restored images and provide image data to remote doctors to support ophthalmic diagnosis.

[0071] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0072] The present invention also provides a storage medium on which a computer program is stored. When the computer program is run by a processor, the above method is executed.

[0073] The storage medium may be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage device, flash memory, magnetic disk or optical disk.

[0074] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A remote optometry diagnosis and treatment intelligent service method, characterized in that: The following steps are involved: Acquiring a fundus image to be processed and performing standardization processing on the image, wherein the standardization processing includes normalizing the grayscale value of the image and performing geometric correction; Performing edge detection processing on the standardized image to extract regions containing structural features in the image, and dividing the image into sub-regions based on the structural feature regions; Calculating information entropy values ​​for the image sub-regions respectively, and generating a region weight map based on the information entropy values ​​for weighted configuration of compression processing for different image sub-regions; Performing feature encoding on the standardized image based on a variational autoencoder, and using the region weight map to adjust the encoder's generation of a latent space representation of each image subregion to generate corresponding latent coding data; The latent coding data is used as input for decoding processing, and image features of the image at different scales are extracted through a multi-scale decomposition method. The image is then subjected to variational decoding and reconstruction based on the image features at each scale to generate an image reconstruction result.

2. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The normalization processing and geometric correction specifically include: Normalizing the grayscale value of the fundus image so that the pixel value range of the image is mapped to the specified grayscale range; Geometric correction involves using affine transformation and perspective transformation to correct the distortion in the image caused by shooting angle and deformation.

3. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The structural feature areas specifically include: The optic disc, macula, and blood vessel areas in fundus images contain high texture information; The extraction of the structural feature region adopts an edge detection algorithm to accurately extract key edge and contour information in the image.

4. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The information entropy value specifically includes: Calculate the entropy value of each image sub-region. The entropy value calculation formula is: ； in, is the entropy value of the image area; is the pixel value in the image area The probability distribution of is the number of pixel values; The first Pixel value; Regions with high entropy values ​​contain more information; In the process of calculating the information entropy value, the probability distribution of each region is calculated through the grayscale histogram.

5. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The regional weight map specifically includes: Generate a region weight map based on the entropy value of each image sub-region; The regional weight map is generated in the following way: ； in, For location The regional weight value at ; Image area The entropy value of is the maximum entropy value of all regions in the image; By using the region weight map, different compression ratios are applied to different regions in the subsequent compression step.

6. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The feature coding specifically includes: Encoding the standardized image based on a variational autoencoder, wherein the encoding process includes extracting deep features of the image through a convolutional neural network and generating a latent space representation through a fully connected layer; During the feature encoding process, a variational inference method based on reparameterization techniques is used to optimize the distribution of the latent space.

7. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The potential coding data specifically includes: The latent space representation data generated in the encoding step represents low-dimensional features of the fundus image, and the feature encoding data is composed of multiple encoding vectors; The latent coding data is further quantized to convert it into discrete data of fixed size; The latent coded data is obtained by adopting the latent space of variational autoencoder To represent it, including the two parameters of mean and variance, it is used to reconstruct the probability distribution of the input image.

8. The remote ophthalmology diagnosis and treatment intelligent service method according to claim 1, characterized in that: The multi-scale decomposition method includes: Perform multi-scale wavelet transform on the input image, decompose the image into multiple frequency components, and encode, decode and reconstruct these frequency components respectively; The multi-scale decomposition method extracts information of the image at different scales; The multi-scale decomposition adopts discrete wavelet transform and processes the low-frequency part and the high-frequency part of the image respectively.

9. A remote ophthalmology and optometry diagnosis and treatment intelligent service system, according to the remote ophthalmology and optometry diagnosis and treatment intelligent service method according to any one of claims 1 to 8, characterized in that: include: Image acquisition and processing module, used to acquire fundus images and perform image standardization, identify key areas and perform weighted processing on the areas according to entropy values; The encoding and transmission module encodes the weighted fundus image based on the variational autoencoder and transmits the compressed data to the remote end; The decoding and recovery module is used to decode the compressed data received at the remote end and restore image details using multi-scale variational reconstruction technology; The evaluation and diagnosis module is used to evaluate the quality of the restored images and provide image data to remote doctors to support ophthalmic diagnosis.

10. A storage medium, characterized in that: A computer program is stored, which, when executed by a computer, implements the remote ophthalmology diagnosis and treatment intelligent service method according to any one of claims 1 to 8.

Images