A deep learning-based adjustable image optimization system and method

By building an adjustable image optimization system for deep learning models, using random optimization and training data sets, the problem that deep learning models cannot optimize different degrees based on external parameters during image optimization is solved, and flexible image optimization processing and efficiency improvement are achieved .

CN114693552BActive Publication Date: 2024-12-17SHANGHAI SOUNDWISE TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210303343.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-03-25
Publication Date
2024-12-17
Estimated Expiration
2042-03-25

AI Technical Summary

Technical Problem

Existing deep learning models cannot optimize different degrees based on external input parameters during image optimization, and lack the ability to process parameters outside the image.

Method used

By constructing an adjustable image optimization system based on deep learning, the deep learning model is pre-trained, and the original data set is randomly optimized using collection, configuration, preprocessing and training submodules to form a parameter set and use it as a training data set. The model is trained to realize multi-class optimization processing of the optimized image, including optimization of smoothing degree, feature intensity, edge sharpness and connection line degree.

Benefits of technology

It realizes image optimization to different degrees according to input parameters, obtains different optimization results of the same image, determines the optimal optimization parameters, and is suitable for optimization research of different images, improving the flexibility and efficiency of deep learning models.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114693552B_ABST
    Figure CN114693552B_ABST
Patent Text Reader

Abstract

The present invention provides an adjustable image optimization system and method based on deep learning, which collects several images as original data sets; configures several types of optimization parameters and several parameter values ​​of each type of optimization parameters; performs random optimization for each image in the original data set for a preset number of times, and randomly selects a parameter value for each optimization parameter during each random optimization to form a parameter set and use it as an input parameter for optimization, so as to obtain the optimized images of each image for a preset number of times; uses the original data set, the optimized image and the corresponding parameter set as training data sets to train the constructed deep learning model; optimizes the image to be optimized based on the parameter values ​​of the selected optimization parameters and the trained deep learning model to obtain the optimized image of the image to be optimized. The optimization of the image to be optimized by the deep learning model will be optimized to different degrees according to the different input optimization parameters.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of image processing technology and ultrasound technology, and in particular to an adjustable image optimization system and method based on deep learning. Background Art

[0002] When using traditional image algorithms to optimize ultrasound images, it is generally possible to configure some parameters regarding the degree of image optimization. However, in algorithms that use deep learning for image optimization, the general deep learning model itself is treated as a black box. It cannot handle parameters other than images very well, and therefore cannot set some optimization parameters accordingly. Deep learning image optimization rarely achieves different degrees of optimization based on the input parameters. Summary of the invention

[0003] Based on the above-mentioned deficiencies, the present invention provides an adjustable image optimization system and method based on deep learning, aiming to solve technical problems in the prior art that the deep learning model cannot optimize the image to different degrees according to external input parameters when optimizing the image.

[0004] A deep learning-based adjustable image optimization system, comprising:

[0005] A model building module, used for pre-training to form a deep learning model for image optimization, the model building module comprising:

[0006] A collection submodule, used to collect several images as original data sets;

[0007] The configuration submodule is used to configure several types of optimization parameters and several parameter values ​​of each type of optimization parameters;

[0008] A preprocessing submodule, connected to the collection submodule and the configuration submodule, for performing a preset number of random optimizations on each of the images in the original data set, and for randomly selecting a parameter value for each type of the optimization parameter during each random optimization to form a parameter set as an input parameter for the random optimization, so as to obtain the preset number of optimized images for each of the images;

[0009] A training submodule, connected to the acquisition submodule and the preprocessing submodule respectively, for training the constructed deep learning model by using the original data set, the optimized image and the corresponding parameter set as training data sets;

[0010] An image optimization module, connected to the model building module, is used to optimize the image to be optimized using the trained deep learning model, and the image optimization module includes:

[0011] A collection submodule, used for collecting images to be optimized;

[0012] A parameter submodule, used for selecting a parameter value of each of the optimization parameters;

[0013] The optimization submodule is respectively connected to the acquisition submodule and the parameter submodule, and is used to optimize the image to be optimized based on the parameter values ​​of the selected optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

[0014] Furthermore, the types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

[0015] Furthermore, each parameter value included in the parameter value of each type of the optimization parameter is used to represent a unique optimization degree of the random optimization.

[0016] Furthermore, an image optimization algorithm is used to randomly optimize each image in the original data set. In the image optimization algorithm, different processing methods are performed for each type of optimization parameters:

[0017] A Gaussian blur algorithm is used according to the degree of smoothness;

[0018] Retinex algorithm is used for the feature strength;

[0019] A Laplace enhancement algorithm is used for the edge sharpness strength;

[0020] The Hough straight line algorithm is used for the connection line degree.

[0021] Furthermore, the deep learning model is a STGAN optimization model.

[0022] Furthermore, the acquisition submodule 5 is arranged in an ultrasonic device.

[0023] Furthermore, the preset number of times is 8.

[0024] A method for adjusting image optimization based on deep learning, characterized in that the method adopts the aforementioned adjustable image optimization system based on deep learning, comprising:

[0025] Steps to pre-train a deep learning model for image optimization:

[0026] Step A1, collecting a number of images as an original data set;

[0027] Step A2, configuring several types of optimization parameters and several parameter values ​​of each type of optimization parameters;

[0028] Step A3, performing a preset number of random optimizations on each of the images in the original data set, randomly selecting a parameter value for each type of the optimization parameter during each random optimization to form a parameter set as an input parameter for the random optimization, so as to obtain the preset number of optimized images for each of the images;

[0029] Step A4, using the original data set, the optimized image and the corresponding parameter set as training data sets to train the constructed deep learning model;

[0030] Steps of optimizing the image to be optimized using the trained deep learning model:

[0031] Step B1, collecting the image to be optimized;

[0032] Step B2, selecting a parameter value for each of the optimization parameters;

[0033] Step B3: optimizing the image to be optimized based on the selected parameter values ​​of the optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

[0034] Furthermore, the types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

[0035] Furthermore, each parameter value included in each type of the optimization parameters is used to represent a unique optimization degree of the random optimization.

[0036] The beneficial technical effect of the present invention is that by pre-optimizing the image to different degrees under different optimization parameters, the parameters and the optimized image are used as input to train the deep learning model. The trained deep learning model will optimize the image to be optimized to different degrees according to different input parameters. Different optimization results for the same image can be obtained, and the best optimization parameters can be obtained in turn. Different images can also be optimized with the same optimization parameters to study the optimization results of different images. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a model building module of an adjustable image optimization system based on deep learning of the present invention;

[0038] Figure 2 A schematic diagram of an image optimization module of an adjustable image optimization system based on deep learning according to the present invention;

[0039] Figure 3 A flowchart of the model construction steps of an adjustable image optimization method based on deep learning of the present invention;

[0040] Figure 4 This is a flow chart of the image optimization steps of an adjustable image optimization method based on deep learning in the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0042] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0043] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments, but they are not intended to limit the present invention.

[0044] See also Figure 1 The present invention provides an adjustable image optimization system based on deep learning, comprising:

[0045] A model building module is used for pre-training to form a deep learning model for image optimization. The model building module includes:

[0046] A collection submodule 1 is used to collect a number of images as an original data set;

[0047] Configuration submodule 2, used to configure several types of optimization parameters and several parameter values ​​of each type of optimization parameters;

[0048] The preprocessing submodule 3 is connected to the collection submodule 1 and the configuration submodule 2, and is used to perform a preset number of random optimizations on each image in the original data set. During each random optimization, a parameter value is randomly selected for each optimization parameter to form a parameter set as an input parameter for the optimization, so as to obtain a preset number of optimized images for each image.

[0049] The training submodule 4 is connected to the acquisition submodule 1 and the preprocessing submodule 3 respectively, and is used to train the constructed deep learning model by using the original data set, the optimized image and the corresponding parameter set as the training data set;

[0050] The image optimization module is connected to the model building module and is used to optimize the image to be optimized using the trained deep learning model. The image optimization module includes:

[0051] The acquisition submodule 5 is used to acquire the image to be optimized;

[0052] Parameter submodule 6, used to select the parameter value of each optimization parameter;

[0053] The optimization submodule 7 is respectively connected to the acquisition submodule 5 and the parameter submodule 6, and is used to optimize the image to be optimized based on the parameter values ​​of the selected optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

[0054] Furthermore, the preset number of times is 8.

[0055] Furthermore, the types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

[0056] Furthermore, each parameter value included in the parameter value of each type of optimization parameter is used to represent a unique optimization degree of the random optimization, including from no optimization degree to complete optimization degree, which is divided into 10 parameter values ​​in total.

[0057] Furthermore, an image optimization algorithm is used to randomly optimize each image in the original data set. In the image optimization algorithm, different processing methods are performed for each type of optimization parameters:

[0058] A Gaussian blur algorithm is used according to the degree of smoothness;

[0059] Retinex algorithm is used for the feature strength;

[0060] A Laplace enhancement algorithm is used for the edge sharpness strength;

[0061] The Hough straight line algorithm is used for the connection line degree.

[0062] Furthermore, the deep learning model is a STGAN optimization model.

[0063] By migrating STGAN to the direction of image optimization and combining it with traditional image optimization algorithms, a deep learning model that can respond to optimization parameters is constructed.

[0064] The acquisition submodule 5 is arranged in an ultrasonic device.

[0065] Specifically, the images used as the original data set are ultrasound images.

[0066] Specifically, the ultrasound image is an image of the thyroid gland.

[0067] Specifically, assuming that there are n images in the original data set, the present invention constructs a set of traditional image optimization algorithms Op through opencv. The input optimization parameters of the image optimization algorithm include smoothness, feature strength, edge sharpness, and connecting line strength. Each optimization parameter is set to 10 parameter values ​​from 0 to 9, that is, 10 levels. Parameter value 0 represents the degree of non-optimization, and parameter value 9 represents the degree of complete optimization. For each image in the original data set, the image optimization algorithm is used for random optimization to obtain an optimized image. Each random optimization randomly selects parameter values ​​of 4 optimization parameters to form a parameter set. For example, a randomly selected parameter set is [4,5,3,8], representing the parameter value of smoothness is 4, the parameter value of feature strength is 5, the parameter value of edge sharpness is 3, and the parameter value of connecting line strength is 8. If each image is randomly optimized 8 times, 8n optimized images will be obtained after the original data set is optimized, and the corresponding number of parameter sets is also 8n. The 8n optimized images, 8n parameter sets and the original data set are used to construct a training data set, and a deep learning model such as an STGAN optimization model is trained.

[0068] After obtaining the trained deep learning model, collect ultrasound images and set the parameter set. For example, if you want a better smoothness, you can set the parameter value of the optimization parameter of the smoothness in the parameter set higher. The parameter set formed by the optimized parameters with the set parameter values ​​and the ultrasound image are used as inputs. After being processed by the deep learning model, the optimized image of the ultrasound image is obtained. Image optimization in response to the optimization parameters is achieved.

[0069] According to the deep learning model trained by the present invention, when the deep learning model is used for image optimization, different optimization effects can be achieved for the same deep learning model, and the different optimization effects depend entirely on the parameter values ​​of the customized different optimization parameters, which is more efficient than other deep learning model optimization methods. Through an adjustable image optimization system based on deep learning of the present invention, using the acquired deep learning model, different parameter sets can be selected to achieve optimization in various aspects for the same image to be optimized, and different optimized images can be obtained, which can be used to study the parameter values ​​of the optimal optimization parameters required for the same image, such as the same medical lesion. For different images, the same parameter set can be used to optimize different images, so as to study the performance of a single optimization parameter in different images, such as different medical lesions.

[0070] Specifically, the pytorch framework is used to train the STGAN optimization model.

[0071] Deep learning models can be deployed in ultrasound devices.

[0072] Furthermore, the acquisition submodule 5 is an ultrasonic device.

[0073] join Figure 2 The present invention also provides an adjustable image optimization method based on deep learning, which adopts the aforementioned adjustable image optimization system based on deep learning, including:

[0074] Steps to pre-train a deep learning model for image optimization:

[0075] Step A1, collecting a number of images as an original data set;

[0076] Step A2, configuring several types of optimization parameters and several parameter values ​​of each type of optimization parameters;

[0077] Step A3, performing a preset number of random optimizations on each image in the original data set, randomly selecting a parameter value for each optimization parameter during each random optimization to form a parameter set as an input parameter for optimization, so as to obtain a preset number of optimized images for each image;

[0078] Step A4, using the original data set, the optimized image and the corresponding parameter set as training data sets to train the constructed deep learning model;

[0079] Steps to optimize the image using the trained deep learning model:

[0080] Step B1, collecting the image to be optimized;

[0081] Step B2, selecting a parameter value for each optimization parameter;

[0082] Step B3, optimizing the image to be optimized based on the parameter values ​​of the selected optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

[0083] Furthermore, the types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

[0084] Furthermore, each parameter value included in the parameter value of each type of optimization parameter is used to represent a unique optimization degree of the random optimization, from the degree of no optimization to the degree of complete optimization, which is divided into 10 parameter values ​​in total.

[0085] The above are only preferred embodiments of the present invention, and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should be aware that all solutions obtained by equivalent substitutions and obvious changes made using the description and illustrations of the present invention should be included in the protection scope of the present invention.

Claims

1. An adjustable image optimization system based on deep learning, characterized in that: include: A model building module, used for pre-training to form a deep learning model for image optimization, the model building module comprising: A collection submodule, used to collect several images as original data sets; The configuration submodule is used to configure several types of optimization parameters and several parameter values ​​of each type of optimization parameters; A preprocessing submodule, connected to the collection submodule and the configuration submodule, for performing a preset number of random optimizations on each of the images in the original data set, and for randomly selecting a parameter value for each type of the optimization parameter each time the random optimization is performed to form a parameter set as an input parameter for the random optimization, so as to obtain the preset number of optimized images for each of the images; A training submodule, connected to the collection submodule and the preprocessing submodule respectively, for training the constructed deep learning model by using the original data set, the optimized image and the corresponding parameter set as training data sets; An image optimization module, connected to the model building module, is used to optimize the image to be optimized using the trained deep learning model, and the image optimization module includes: A collection submodule, used for collecting images to be optimized; A parameter submodule, used for selecting a parameter value of each of the optimization parameters; The optimization submodule is respectively connected to the acquisition submodule and the parameter submodule, and is used to optimize the image to be optimized based on the parameter values ​​of the selected optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

2. The deep learning-based adjustable image optimization system according to claim 1, characterized in that: The types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

3. The deep learning-based adjustable image optimization system according to claim 2, characterized in that: Each parameter value included in the parameter value of each type of the optimization parameter is used to represent a unique optimization degree of the random optimization.

4. The deep learning-based adjustable image optimization system according to claim 2, characterized in that: An image optimization algorithm is used to randomly optimize each image in the original data set. In the image optimization algorithm, different processing methods are performed for each type of optimization parameters: A Gaussian blur algorithm is used according to the degree of smoothness; Retinex algorithm is used for the feature strength; A Laplace enhancement algorithm is used for the edge sharpness strength; The Hough straight line algorithm is used for the connection line degree.

5. The deep learning-based adjustable image optimization system according to claim 1, characterized in that: The deep learning model is a STGAN optimization model.

6. The deep learning-based adjustable image optimization system according to claim 1, characterized in that: The acquisition submodule is arranged in an ultrasonic device.

7. The deep learning-based adjustable image optimization system according to claim 1, characterized in that: The preset number of times is 8.

8. An adjustable image optimization method based on deep learning, characterized in that: A deep learning-based adjustable image optimization system as described in any one of claims 1 to 7 is used, comprising: Steps to pre-train a deep learning model for image optimization: Step A1, collecting a number of images as an original data set; Step A2, configuring several types of optimization parameters and several parameter values ​​of each type of optimization parameters; Step A3, performing a preset number of random optimizations on each of the images in the original data set, randomly selecting a parameter value for each type of the optimization parameter during each random optimization to form a parameter set as an input parameter for the random optimization, so as to obtain the preset number of optimized images for each of the images; Step A4, using the original data set, the optimized image and the corresponding parameter set as training data sets to train the constructed deep learning model; Steps of optimizing the image to be optimized using the trained deep learning model: Step B1, collecting the image to be optimized; Step B2, selecting a parameter value for each of the optimization parameters; Step B3: optimizing the image to be optimized based on the selected parameter values ​​of the optimization parameters and the trained deep learning model to obtain an optimized image of the image to be optimized.

9. The adjustable image optimization method based on deep learning as claimed in claim 8, characterized in that: The types of optimization parameters include smoothness, feature strength, edge sharpness, and connection line degree.

10. The adjustable image optimization method based on deep learning according to claim 9, characterized in that: Each parameter value included in each type of the optimization parameters is used to represent a unique optimization degree of the random optimization.