A method and system for infrared image optimization based on deep learning

Abstract

The present invention provides an infrared image optimization method and system based on deep learning, including: performing a target number of non-uniform noise iterative superpositions on an initial infrared image, and independently recording each non-uniform noise superposition result to construct an infrared image training set; performing deep learning on the infrared image training set based on a deep neural network to construct a diffusion model, and updating and optimizing the model parameters of the diffusion model based on the measured data of the diffusion model; deploying the updated and optimized diffusion model, and performing infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result to complete the optimization of the infrared image. The infrared image processing effect is guaranteed while the processing efficiency is improved, and the detailed features of the infrared image are retained while the infrared image is denoised, effectively balancing the relationship between denoising and image details, and ensuring the optimization reliability of the infrared image.

Description

Technical Field

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

[0002] Infrared image optimization is infrared image denoising. Infrared image denoising refers to the process of removing interfering noise caused by sensor noise, signal interference or other factors in the image through a series of processing in digital image processing to improve the quality and clarity of the image. The purpose is to reduce the impact of noise in the image and make the image clearer and accurately reflect the original scene.

[0003] However, the current infrared image denoising has the following disadvantages:

[0004] (1) Information loss: Some denoising algorithms may cause loss of image information. In particular, important details or features in the image may be lost while reducing noise, resulting in image distortion.

[0005] (2) Edge blur: Some denoising algorithms can cause image edge blur, which affects the accuracy of target detection and recognition and may cause misjudgment and false detection;

[0006] (3) Poor adaptability: Some denoising algorithms have poor adaptability to infrared images in different scenes and need to be adjusted for specific types of noise and image characteristics. If not handled properly, the effect may be poor;

[0007] (4) Difficult to strike a balance: In denoising, reducing noise while retaining image details and features requires a good balance, and currently it is impossible to accurately balance the relationship between the two;

[0008] Therefore, in order to overcome the above-mentioned defects, the present invention provides an infrared image optimization method and system based on deep learning. Summary of the invention

[0009] The present invention provides an infrared image optimization method and system based on deep learning, which are used to accurately and effectively construct an infrared image training set by iteratively superimposing non-uniform noise on an initial infrared image for a target number of times and recording each superposition result. Secondly, the infrared image training set obtained is learned and trained by a deep neural network to accurately and reliably construct a diffusion model, and the diffusion model is continuously updated and optimized according to test data of the diffusion model, thereby providing convenience and guarantee for optimizing the infrared image and ensuring the accuracy and reliability of the diffusion model finally obtained. Finally, the obtained diffusion model is deployed, so as to facilitate fast and effective processing of different infrared images to be processed according to the deployment results, thereby ensuring the processing effect of the infrared image while improving the processing efficiency, retaining the detailed features of the infrared image while denoising the infrared image, effectively balancing the relationship between denoising and image details, and ensuring the optimization reliability of the infrared image.

[0010] The present invention provides an infrared image optimization method based on deep learning, comprising:

[0011] Perform a target number of non-uniform noise iterative superpositions on the initial infrared image, and independently record each non-uniform noise superposition result to construct an infrared image training set;

[0012] Based on the deep neural network, the infrared image training set is deeply learned to build a diffusion model, and the model parameters of the diffusion model are updated and optimized based on the measured data of the diffusion model;

[0013] The diffusion model obtained by updating and optimizing is deployed, and based on the deployment result, infrared image denoising and enhancement processing is performed on the infrared image to be processed to complete the optimization of the infrared image.

[0014] Preferably, a deep learning-based infrared image optimization method, before performing a target number of non-uniform noise iterative superposition on an initial infrared image, comprises:

[0015] Access the image library based on the server, and retrieve the record log of the image library based on the access result;

[0016] Parse the record logs to determine the basic storage parameters of the infrared images in the image library, and determine the extraction scheme for the infrared images based on the basic storage parameters;

[0017] The infrared image in the image library is extracted based on the extraction scheme, and the initial infrared image is obtained based on the extraction result.

[0018] Preferably, a deep learning-based infrared image optimization method performs a target number of non-uniform noise iterative superpositions on an initial infrared image, comprising:

[0019] Acquire the obtained initial infrared image, and determine a difference noise superposition scheme for different initial infrared images based on image features of the initial infrared image;

[0020] Determine available noise of different initial infrared images based on a difference noise superposition scheme, and generate noise images of the same size according to size parameters of the initial infrared images based on the available noise;

[0021] Based on the noise superposition requirement, the weight parameter of the noise image when iteratively superimposing the non-uniform noise with the initial infrared image is determined;

[0022] The noise image is adaptively enhanced based on the weight parameter, and the adaptively enhanced noise image is iteratively superimposed with the initial infrared image for a target number of non-uniform noises.

[0023] Preferably, a method for optimizing infrared images based on deep learning is to iteratively superimpose a target number of non-uniform noises on the noise image after adaptive enhancement processing and the initial infrared image, including:

[0024] Performing foreground and background separation on the initial infrared image, and determining image adjustment variable values ​​for the initial infrared image based on the foreground and background separation result;

[0025] Determining the retention features of the initial infrared image and the image contrast based on the image adjustment variable values;

[0026] The noise image and the initial infrared image are converted into an image matrix based on the preserved features and image contrast, and the noise image and the initial infrared image are added based on the image matrix to complete the iterative superposition of non-uniform noise.

[0027] Preferably, an infrared image optimization method based on deep learning independently records each non-uniform noise superposition result to construct an infrared image training set, including:

[0028] An image version monitoring thread is constructed based on the state change parameters of the initial infrared image, and the non-uniform noise superposition result of the initial infrared image is fully monitored based on the image version monitoring thread;

[0029] Construct a record table, and determine the basic parameters corresponding to each non-uniform noise superposition result and the processed infrared image corresponding to each non-uniform noise superposition result based on the full monitoring results, and record the basic parameters in sequence in the record table;

[0030] The processed infrared image corresponding to each non-uniform noise superposition result is converted into a separate file, and the separate file is associated with the corresponding basic parameters in the record table to obtain an infrared image training set.

[0031] Preferably, a deep learning-based infrared image optimization method performs deep learning on an infrared image training set based on a deep neural network to construct a diffusion model, including:

[0032] Acquire the infrared image training set, and separate the image contour and the infrared object image of each infrared image in the infrared image training set;

[0033] Based on the deep learning network, the first target thread and the second target thread are respectively called according to the splitting result to process the image contour and the infrared object image according to the preset pixel gradient feature calculation kernel function to obtain the contour gradient feature and the infrared object gradient feature;

[0034] Summarizing the contour gradient features and the infrared object gradient features to obtain the change rate of the infrared image in different directions, and determining the image gradient of the infrared image based on the change rate;

[0035] Based on the image gradient, the balance parameters of denoising different types of infrared images and retaining infrared image details are determined, and the diffusion coefficients of denoising different types of infrared images are determined based on the balance parameters, and the basic equation of difference denoising is constructed based on the diffusion coefficient and the benchmark denoising requirements;

[0036] At the same time, determining the application scenario of the denoised infrared image, and determining the image enhancement dimension of the denoised infrared image based on the application scenario;

[0037] Determine the control variables and hybrid processing schemes of different image enhancement dimensions within the display range of the device based on the visual expression requirements, and formulate the image enhancement processing mechanism for infrared images based on the control variables and hybrid processing schemes;

[0038] The basic equations of difference denoising and image enhancement processing mechanism are learned and trained, and a diffusion model is constructed based on the learning and training structure.

[0039] Preferably, a deep learning-based infrared image optimization method updates and optimizes the model parameters of the diffusion model based on measured data of the diffusion model, comprising:

[0040] Acquire multiple test infrared images, and sequentially input the multiple test infrared images into the diffusion model for processing to obtain measured data of each test infrared image;

[0041] Retrieve the reference infrared optimized image of each test infrared image, and perform image key point feature similarity matching between the reference infrared optimized image and the measured data;

[0042] Determine the difference ratio between the measured data and the reference infrared optimized image based on the matching result, and when the difference ratio exceeds a preset threshold, determine the updated optimization dimension of the diffusion model and the updated optimization parameter on the updated optimization dimension based on the measured data;

[0043] The model parameters of the diffusion model are updated and optimized based on the update optimization dimension and the update optimization parameter on the update optimization dimension.

[0044] Preferably, a deep learning-based infrared image optimization method deploys the updated and optimized diffusion model, including:

[0045] Determine the deployment environment of the diffusion model based on the usage scenario of the diffusion model, and determine the environmental adaptation parameters for the deployment environment in the server based on the operation requirements of the diffusion model;

[0046] Adapt the deployment environment based on the environment adaptation parameters, and export the diffusion model based on the environment adaptation result to obtain a diffusion model file;

[0047] Determine the current available space in the server based on the basic operating parameters of the server, and adaptively adjust the file size of the diffusion model file based on the current available space to obtain a deployable diffusion model file;

[0048] Based on the environmental characteristics of the deployment environment, the low-code is logically combined to build the model deployment code, and the model deployment code and the deployable diffusion model file are uploaded to the deployment environment;

[0049] Based on the preset deployment service, the deployable diffusion model is deployed according to the uploaded result, and the interface test is performed on the deployment result. When the interface communication status meets the preset requirements, the deployment of the diffusion model is completed.

[0050] Preferably, a deep learning-based infrared image optimization method performs infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result to complete the optimization of the infrared image, including:

[0051] Based on the deployment results, the diffusion model is multi-dimensionally connected with the working platform, and based on the multi-dimensional connection results, infrared images to be processed uploaded by different working platforms are received;

[0052] The infrared images to be processed are sorted based on the upload timestamps, and the infrared images to be processed are sequentially transmitted to the diffusion model based on the sorting results, and the infrared images to be processed uploaded at the same time are subjected to parallel infrared image denoising and enhancement processing in the deployed diffusion model;

[0053] The parallel infrared image denoising and enhancement processing results are fed back to each working platform in turn to complete the optimization of the infrared image.

[0054] The present invention provides an infrared image optimization system based on deep learning, comprising:

[0055] The training set construction module is used to iteratively superimpose non-uniform noise on the initial infrared image for a target number of times, and independently record each non-uniform noise superposition result to construct an infrared image training set;

[0056] The model building and optimization module is used to perform deep learning on the infrared image training set based on a deep neural network, build a diffusion model, and update and optimize the model parameters of the diffusion model based on the measured data of the diffusion model;

[0057] The infrared image processing module is used to deploy the diffusion model obtained by updating and optimizing, and to perform infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result, so as to complete the optimization of the infrared image.

[0058] Compared with the prior art, the present invention has the following beneficial effects:

[0059] 1. By iteratively superimposing the target number of non-uniform noises on the initial infrared image and recording the results of each superposition, the infrared image training set can be accurately and effectively constructed. Secondly, the infrared image training set is learned and trained by a deep neural network to accurately and reliably construct the diffusion model, and the diffusion model is continuously updated and optimized according to the test data of the diffusion model, which provides convenience and guarantee for the optimization of infrared images and ensures the accuracy and reliability of the final diffusion model. Finally, the obtained diffusion model is deployed to facilitate the fast and effective processing of different infrared images to be processed according to the deployment results, ensuring the processing effect of the infrared image while improving the processing efficiency, and retaining the detailed features of the infrared image while denoising the infrared image, effectively balancing the relationship between denoising and image details, and ensuring the optimization reliability of the infrared image.

[0060] 2. By analyzing the infrared image training set obtained, the image contour and infrared object image in each infrared image are split, so as to facilitate the accurate and effective determination of the image gradient of different areas in the infrared image. Secondly, the image contour and the infrared object image are processed according to the first target thread, the second target thread and the preset pixel gradient feature calculation kernel function, so as to achieve accurate and effective determination of the image gradient of the infrared image. At the same time, the diffusion coefficient of different types of infrared images during denoising is determined by the image gradient, and the basic equation of difference denoising is constructed according to the diffusion coefficient, so as to facilitate effective denoising of different infrared images. Finally, by determining the application scenario of the denoised infrared image, the image enhancement dimension of different infrared images is determined, and the image enhancement processing mechanism of the infrared image is determined according to the image enhancement dimension, and the basic equation of difference denoising and the image enhancement processing mechanism are learned and trained, and finally the diffusion model is accurately and reliably constructed, which provides convenience and guarantee for the optimization of infrared images, ensures the denoising effect of infrared images, effectively balances the relationship between denoising and image details, and improves the accuracy and reliability of infrared image optimization.

[0061] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0062] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0064] Figure 1 is a flow chart of an infrared image optimization method based on deep learning in an embodiment of the present invention;

[0065] Figure 2 This is a flowchart of step 1 in a method for optimizing infrared images based on deep learning in an embodiment of the present invention;

[0066] Figure 3 This is a structural diagram of an infrared image optimization system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0068] Embodiment 1:

[0069] This embodiment provides an infrared image optimization method based on deep learning, such as Figure 1 As shown, including:

[0070] Step 1: Perform a target number of non-uniform noise iterative superpositions on the initial infrared image, and independently record each non-uniform noise superposition result to construct an infrared image training set;

[0071] Step 2: Perform deep learning on the infrared image training set based on a deep neural network to build a diffusion model, and update and optimize the model parameters of the diffusion model based on the measured data of the diffusion model;

[0072] Step 3: Deploy the diffusion model obtained by the update optimization, and perform infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result to complete the optimization of the infrared image.

[0073] In this embodiment, the initial infrared images are known in advance and are infrared images of different types retrieved from an image library and used to construct an infrared image training set.

[0074] In this embodiment, the iterative superposition of non-uniform noise refers to using different types of noise to perform noise processing on the corresponding type of initial infrared image, for example, it can be Gaussian noise, salt and pepper noise, or stripe noise.

[0075] In this embodiment, the target number is known in advance and is used to represent the number of times the initial infrared image is denoised, and the result of each denoising process is different, thereby achieving the construction of an infrared image training set.

[0076] In this embodiment, the deep neural network is known in advance.

[0077] In this embodiment, the diffusion model is constructed by deep learning the infrared image training set through a deep neural network, and is used to simulate the diffusion and thrust process of the infrared image, thereby realizing denoising processing of the infrared image.

[0078] In this embodiment, the measured data refers to the performance data of the diffusion model obtained after a simulation test is performed on the constructed diffusion model.

[0079] In this embodiment, the model parameters refer to the working parameters of each construction part of the diffusion model.

[0080] In this embodiment, the enhancement processing refers to adjusting the contrast, brightness, etc. of the denoised infrared image, so as to facilitate the display of the denoised infrared image.

[0081] The working principle and beneficial effects of the above technical solution are: by iteratively superimposing non-uniform noise on the initial infrared image for a target number of times and recording the results of each superposition, an accurate and effective construction of an infrared image training set is achieved; secondly, the infrared image training set obtained is learned and trained through a deep neural network to achieve accurate and reliable construction of a diffusion model, and the diffusion model is continuously updated and optimized according to the test data of the diffusion model, thereby providing convenience and guarantee for the optimization of infrared images and ensuring the accuracy and reliability of the final diffusion model; finally, the obtained diffusion model is deployed, so as to facilitate fast and effective processing of different infrared images to be processed according to the deployment results, thereby ensuring the processing effect of the infrared image while improving the processing efficiency, and retaining the detailed features of the infrared image while denoising the infrared image, effectively balancing the relationship between denoising and image details, and ensuring the optimization reliability of the infrared image.

[0082] Embodiment 2:

[0083] Based on Example 1, this embodiment provides an infrared image optimization method based on deep learning, which includes:

[0084] Access the image library based on the server, and retrieve the record log of the image library based on the access result;

[0085] Parse the record logs to determine the basic storage parameters of the infrared images in the image library, and determine the extraction scheme for the infrared images based on the basic storage parameters;

[0086] The infrared image in the image library is extracted based on the extraction scheme, and the initial infrared image is obtained based on the extraction result.

[0087] In this embodiment, the image library is pre-set and used to store a variety of different types of images, wherein the infrared image is one of the images in the image library.

[0088] In this embodiment, the record log is used to record parameter information such as the storage location and quantity of different types of images in the image library.

[0089] In this embodiment, the basic storage parameters refer to information such as the storage location of the infrared image in the image library.

[0090] In this embodiment, the extraction scheme refers to the strategy for extracting infrared images from the image library. For example, the infrared images in the image library may be grouped first, and an equal number of infrared images may be extracted from each group, thereby achieving the purpose of obtaining infrared images from the image library.

[0091] The working principle and beneficial effects of the above technical solution are: by parsing the record logs of the image library, the basic storage parameters of the infrared images in the image library can be accurately and effectively acquired, which provides convenience for extracting infrared images; secondly, the obtained basic storage parameters are parsed to accurately and effectively formulate the extraction scheme for the infrared images in the image library; finally, the infrared images are accurately and reliably extracted according to the formulated extraction method, which provides reliable samples for non-uniform noise superposition, ensuring the comprehensiveness and reliability of the final infrared image training set.

[0092] Embodiment 3:

[0093] Based on Example 1, this embodiment provides an infrared image optimization method based on deep learning, such as Figure 2 As shown, the initial infrared image is iteratively superimposed with non-uniform noise for a target number of times, including:

[0094] Step 101: acquiring an initial infrared image, and determining a difference noise superposition scheme for different initial infrared images based on image features of the initial infrared image;

[0095] Step 102: determining available noises of different initial infrared images based on a difference noise superposition scheme, and generating noise images of the same size according to size parameters of the initial infrared images based on the available noises;

[0096] Step 103: determining a weight parameter of the noise image when performing iterative non-uniform noise superposition with the initial infrared image based on the noise superposition requirement;

[0097] Step 103: Adaptively enhance the noise level of the noise image based on the weight parameter, and perform a target number of non-uniform noise iterative superpositions on the adaptively enhanced noise image and the initial infrared image.

[0098] In this embodiment, the image features of the initial infrared image are used to characterize the objects recorded in the initial infrared image and the pixel distribution in the initial infrared image.

[0099] In this embodiment, the difference noise superposition scheme refers to using different noise superposition schemes to perform non-uniform noise superposition on initial infrared images with different image features, such as Gaussian noise superposition, salt and pepper noise superposition, and mixed noise superposition of Gaussian noise and salt and pepper noise.

[0100] In this embodiment, the available noise refers to the noise type applicable to different initial infrared images, for example, it may be one of Gaussian noise, salt and pepper noise, and a mixed noise of Gaussian noise and salt and pepper noise.

[0101] In this embodiment, the noise superposition requirement is known in advance.

[0102] In this embodiment, the weight parameter is used to characterize the degree of each superposition when the noise is superimposed on the initial infrared image through the noise image. The larger the value, the heavier the superimposed noise.

[0103] In this embodiment, the adaptive noise level enhancement process is to adjust the noise level accordingly according to the value of the weight parameter, so as to achieve different degrees of noise superposition on the initial infrared image.

[0104] The working principle and beneficial effects of the above technical scheme are: by accurately and effectively determining the difference noise superposition scheme of different initial infrared images according to the image characteristics of the initial infrared image, and locking the available noise of different initial infrared images according to the difference noise superposition scheme, secondly, generating a noise image of corresponding size according to the available noise and the size parameters of the initial infrared image, and determining the weight parameters for iterative noise superposition of different initial infrared images according to the superposition requirements, so as to facilitate different degrees of noise superposition processing of different initial infrared images, finally, determining the adaptive noise level enhancement processing of the noise image according to the weight parameter, and performing the target number of non-uniform noise iterative superposition on the processed noise image and the initial infrared image, thereby ensuring the effect of noise superposition on the initial infrared image and the comprehensiveness and reliability of the infrared image training set finally obtained.

[0105] Embodiment 4:

[0106] On the basis of Example 3, this embodiment provides an infrared image optimization method based on deep learning, which performs a target number of non-uniform noise iterative superposition on the noise image after adaptive enhancement processing and the initial infrared image, including:

[0107] Performing foreground and background separation on the initial infrared image, and determining image adjustment variable values ​​for the initial infrared image based on the foreground and background separation result;

[0108] Determining the retention features of the initial infrared image and the image contrast based on the image adjustment variable values;

[0109] The noise image and the initial infrared image are converted into an image matrix based on the preserved features and image contrast, and the noise image and the initial infrared image are added based on the image matrix to complete the iterative superposition of non-uniform noise.

[0110] In this embodiment, foreground-background separation refers to distinguishing the background image and the recorded object in the initial infrared image, in order to determine the detail features of the initial infrared image that need to be retained when noise is superimposed and the image contrast corresponding to the initial infrared image.

[0111] In this embodiment, the image matrix refers to data obtained by format conversion of the noise image and the initial infrared image according to the retained features and image contrast, and is a processing format that facilitates superposition of the noise image and the initial infrared image.

[0112] The working principle and beneficial effects of the above technical solution are: by separating the foreground and background of the initial infrared image, the retained features and image contrast of the initial infrared image can be accurately and effectively determined; secondly, the noise image and the initial infrared image are synchronously converted into an image matrix according to the retained features and image contrast, so as to facilitate the noise superposition processing operation on the initial infrared image through the noise image; finally, the image matrices corresponding to the noise image and the initial infrared image are added to achieve accurate and reliable non-uniform noise iterative superposition of the initial infrared image, thereby ensuring the effect of noise superposition.

[0113] Embodiment 5:

[0114] On the basis of Example 1, this embodiment provides an infrared image optimization method based on deep learning, which independently records each non-uniform noise superposition result and constructs an infrared image training set, including:

[0115] An image version monitoring thread is constructed based on the state change parameters of the initial infrared image, and the non-uniform noise superposition result of the initial infrared image is fully monitored based on the image version monitoring thread;

[0116] Construct a record table, and determine the basic parameters corresponding to each non-uniform noise superposition result and the processed infrared image corresponding to each non-uniform noise superposition result based on the full monitoring results, and record the basic parameters in sequence in the record table;

[0117] The processed infrared image corresponding to each non-uniform noise superposition result is converted into a separate file, and the separate file is associated with the corresponding basic parameters in the record table to obtain an infrared image training set.

[0118] In this embodiment, the state change parameter refers to the degree of change in the state of the initial infrared image. The purpose is to construct an image version monitoring thread, so as to automatically record the noise superposition result of each initial infrared image, and ensure the accuracy and reliability of the final infrared image training set. Among them, the image version monitoring thread is to timely monitor the result of each noise superposition, so as to realize the accurate and effective construction of the infrared image training set.

[0119] In this embodiment, the basic parameters refer to image parameters of the initial infrared image after each noise superposition when noise is superimposed on the initial infrared image, including pixel values, resolution values, etc.

[0120] In this embodiment, processing the infrared image refers to obtaining the infrared image after each noise superposition.

[0121] The working principle and beneficial effects of the above technical solution are: by constructing an image version monitoring thread according to the state change of the initial infrared image, the non-uniform noise superposition of the initial infrared image can be monitored in real time through the image version monitoring thread, so as to facilitate the accurate and effective acquisition and recording of each noise superposition result. Finally, the monitoring results are recorded to achieve accurate and effective construction of the infrared image training set, which provides reliable data support for the construction of the diffusion model, thereby providing convenience and guarantee for the optimization of the infrared image.

[0122] Embodiment 6:

[0123] On the basis of Example 1, this embodiment provides an infrared image optimization method based on deep learning, which performs deep learning on an infrared image training set based on a deep neural network to construct a diffusion model, including:

[0124] Acquire the infrared image training set, and separate the image contour and the infrared object image of each infrared image in the infrared image training set;

[0125] Based on the deep learning network, the first target thread and the second target thread are respectively called according to the splitting result to process the image contour and the infrared object image according to the preset pixel gradient feature calculation kernel function to obtain the contour gradient feature and the infrared object gradient feature;

[0126] Summarizing the contour gradient features and the infrared object gradient features to obtain the change rate of the infrared image in different directions, and determining the image gradient of the infrared image based on the change rate;

[0127] Based on the image gradient, the balance parameters of denoising different types of infrared images and retaining infrared image details are determined, and the diffusion coefficients of denoising different types of infrared images are determined based on the balance parameters, and the basic equation of difference denoising is constructed based on the diffusion coefficient and the benchmark denoising requirements;

[0128] At the same time, determining the application scenario of the denoised infrared image, and determining the image enhancement dimension of the denoised infrared image based on the application scenario;

[0129] Determine the control variables and hybrid processing schemes of different image enhancement dimensions within the display range of the device based on the visual expression requirements, and formulate the image enhancement processing mechanism for infrared images based on the control variables and hybrid processing schemes;

[0130] The basic equations of difference denoising and image enhancement processing mechanism are learned and trained, and a diffusion model is constructed based on the learning and training structure.

[0131] In this embodiment, the infrared object image refers to an object recorded in the infrared image, for example, it may be a facial image of a person or an image of an object.

[0132] In this embodiment, the first target thread and the second target thread are set in advance and are used to process the image contour and the infrared object image respectively, that is, to determine the gradient features of the image contour and the infrared object image respectively.

[0133] In this embodiment, the preset pixel gradient feature calculation kernel function is set in advance and is used to determine the gradient of the image.

[0134] In this embodiment, the contour gradient feature refers to the spatial change rate or gradient of the grayscale value in the infrared image contour, and is used to represent the grayscale change degree of each pixel in the infrared image.

[0135] In this embodiment, the object gradient feature refers to the spatial change rate or gradient of the grayscale value in the infrared object image region in the infrared image.

[0136] In this embodiment, the image gradient refers to a feature obtained by summarizing the obtained contour gradient feature and the infrared object gradient feature, and is used to characterize the change of the grayscale value of the entire infrared image in different directions.

[0137] In this embodiment, the balance parameter refers to balancing the denoising degree and the image detail retention degree when denoising different types of infrared images, in order to ensure the denoising effect while retaining more image detail features.

[0138] In this embodiment, the diffusion coefficient is used to adjust or control the intensity or rate of the image diffusion process to achieve a specific image processing effect.

[0139] In this embodiment, the baseline denoising requirement is known in advance and is used to characterize the degree of denoising, etc.

[0140] In this embodiment, the difference denoising basic equation is constructed by using the diffusion coefficient and the basic denoising requirement, and is used to perform denoising processing on the infrared image.

[0141] In this embodiment, the image enhancement dimension refers to an item for performing image enhancement on the infrared image after denoising, such as color rendering.

[0142] In this embodiment, the control variable refers to the parameter size that can adjust the image parameter when performing image enhancement processing on the infrared image, that is, adaptive adjustment is performed within an allowable range.

[0143] In this embodiment, the hybrid processing solution refers to performing image enhancement processing of different items on the denoised infrared image by mixing image enhancement mechanisms of different dimensions.

[0144] The working principle and beneficial effects of the above technical solution are as follows: by analyzing the obtained infrared image training set, the image contour and the infrared object image in each infrared image are split, so as to facilitate the accurate and effective determination of the image gradient of different areas in the infrared image; secondly, by processing the image contour and the infrared object image according to the first target thread, the second target thread and the preset pixel gradient feature calculation kernel function, the image gradient of the infrared image is accurately and effectively determined; at the same time, the diffusion coefficient of different types of infrared images during denoising is determined by the image gradient, and the basic equation of difference denoising is constructed according to the diffusion coefficient, so as to facilitate the effective denoising of different infrared images; finally, by determining the application scenario of the denoised infrared image, the image enhancement dimension of different infrared images is determined, the image enhancement processing mechanism of the infrared image is determined according to the image enhancement dimension, and the basic equation of difference denoising and the image enhancement processing mechanism are learned and trained, and finally the diffusion model is accurately and reliably constructed, which provides convenience and guarantee for the optimization of infrared images, ensures the denoising effect of infrared images, effectively balances the relationship between denoising and image details, and improves the accuracy and reliability of infrared image optimization.

[0145] Embodiment 7:

[0146] On the basis of Example 1, this embodiment provides an infrared image optimization method based on deep learning, which updates and optimizes the model parameters of the diffusion model based on the measured data of the diffusion model, including:

[0147] Acquire multiple test infrared images, and sequentially input the multiple test infrared images into the diffusion model for processing to obtain measured data of each test infrared image;

[0148] Retrieve the reference infrared optimized image of each test infrared image, and perform image key point feature similarity matching between the reference infrared optimized image and the measured data;

[0149] Determine the difference ratio between the measured data and the reference infrared optimized image based on the matching result, and when the difference ratio exceeds a preset threshold, determine the updated optimization dimension of the diffusion model and the updated optimization parameter on the updated optimization dimension based on the measured data;

[0150] The model parameters of the diffusion model are updated and optimized based on the update optimization dimension and the update optimization parameter on the update optimization dimension.

[0151] In this embodiment, the measured data refers to the analysis results obtained after the diffusion model analyzes and processes a plurality of test infrared images.

[0152] In this embodiment, the reference infrared optimized image refers to the standard processing result corresponding to each test infrared image, that is, the corresponding infrared image when the requirements are met.

[0153] In this embodiment, the key points of the image refer to the key image information corresponding to the reference infrared optimized image and the measured data, and the key image information is not unique, and the key image information of the two corresponds to each other.

[0154] In this embodiment, the difference ratio refers to the degree of difference between the measured data and the reference infrared optimized image.

[0155] In this embodiment, the preset threshold is set in advance and is used to measure whether the minimum standard of the processing requirement is met, and can be adjusted.

[0156] In this embodiment, the updating optimization dimension refers to the item or parameter type that needs to be optimized.

[0157] In this embodiment, updating the optimization parameter refers to the specific size of the numerical adjustment of the optimized item or parameter type.

[0158] The working principle and beneficial effects of the above technical solution are: by acquiring multiple test infrared images, and inputting the test infrared images into the diffusion model for analysis and processing, at the same time, the obtained measured data is matched with the benchmark infrared optimized image of each test infrared image for similarity, so as to achieve accurate and effective evaluation of the processing effect of the diffusion model according to the similarity matching results, and according to the matching results, the updated optimization dimension and the updated optimization parameters on the updated optimization dimension are locked, which provides direction and basis for the optimization of the diffusion model, thereby ensuring the accuracy and reliability of the final diffusion model.

[0159] Embodiment 8:

[0160] Based on Example 1, this embodiment provides an infrared image optimization method based on deep learning, which deploys the diffusion model obtained by updating and optimizing, including:

[0161] Determine the deployment environment of the diffusion model based on the usage scenario of the diffusion model, and determine the environmental adaptation parameters for the deployment environment in the server based on the operation requirements of the diffusion model;

[0162] Adapt the deployment environment based on the environment adaptation parameters, and export the diffusion model based on the environment adaptation result to obtain a diffusion model file;

[0163] Determine the current available space in the server based on the basic operating parameters of the server, and adaptively adjust the file size of the diffusion model file based on the current available space to obtain a deployable diffusion model file;

[0164] Based on the environmental characteristics of the deployment environment, the low-code is logically combined to build the model deployment code, and the model deployment code and the deployable diffusion model file are uploaded to the deployment environment;

[0165] Based on the preset deployment service, the deployable diffusion model is deployed according to the uploaded result, and the interface test is performed on the deployment result. When the interface communication status meets the preset requirements, the deployment of the diffusion model is completed.

[0166] In this embodiment, the operation requirements are known in advance and are used to characterize the requirements of the diffusion model on the deployment environment during operation.

[0167] In this embodiment, the environment adaptation parameter refers to the adjustment value of each parameter in the deployment environment, and the purpose is to ensure that the diffusion model can run stably in the deployment environment.

[0168] In this embodiment, the diffusion model file refers to a model file obtained by exporting the trained diffusion model, so as to facilitate the deployment of the diffusion model.

[0169] In this embodiment, adaptive file size adjustment refers to compressing or pruning the diffusion model file according to the available space in the server (simplifying the processing process of the diffusion model), so as to achieve effective deployment of the diffusion model in the server.

[0170] The working principle and beneficial effects of the above technical solution are: by determining the deployment environment of the diffusion model and adapting the deployment environment in the server according to the operation requirements, it is ensured that the diffusion model can run stably in the server; secondly, by determining the current available space of the server, the diffusion model file is adaptively resized according to the current available space to ensure that the diffusion model can be effectively deployed in the server; finally, the interface test of the deployed human diffusion model is carried out, and when the communication requirements are met, the deployment of the diffusion model is completed, ensuring the reliability of the deployment of the diffusion model.

[0171] Embodiment 9:

[0172] Based on Example 1, this embodiment provides an infrared image optimization method based on deep learning, which performs infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result to complete the optimization of the infrared image, including:

[0173] Based on the deployment results, the diffusion model is multi-dimensionally connected with the working platform, and based on the multi-dimensional connection results, infrared images to be processed uploaded by different working platforms are received;

[0174] The infrared images to be processed are sorted based on the upload timestamps, and the infrared images to be processed are sequentially transmitted to the diffusion model based on the sorting results, and the infrared images to be processed uploaded at the same time are subjected to parallel infrared image denoising and enhancement processing in the deployed diffusion model;

[0175] The parallel infrared image denoising and enhancement processing results are fed back to each working platform in turn to complete the optimization of the infrared image.

[0176] In this embodiment, multi-dimensional docking refers to docking the deployed diffusion model with different working platforms simultaneously, so as to enable infrared images generated by different working platforms to be transmitted to the diffusion model in a timely manner for processing.

[0177] In this embodiment, the infrared images to be processed refer to infrared images generated during the operation of different working platforms.

[0178] In this embodiment, the upload timestamp refers to the specific time information when different working platforms transmit the infrared image to be processed to the diffusion model.

[0179] The working principle and beneficial effect of the above technical solution are: by multi-dimensionally connecting the deployed diffusion model with different working platforms, the unprocessed infrared images generated by different working platforms can be timely and effectively transmitted to the diffusion model for denoising and enhancement processing, thereby realizing automatic optimization processing of infrared images and ensuring the effect and reliability of infrared image optimization.

[0180] Embodiment 10:

[0181] This embodiment provides an infrared image optimization system based on deep learning, such as Figure 3 As shown, including:

[0182] The training set construction module is used to iteratively superimpose non-uniform noise on the initial infrared image for a target number of times, and independently record each non-uniform noise superposition result to construct an infrared image training set;

[0183] The model building and optimization module is used to perform deep learning on the infrared image training set based on a deep neural network, build a diffusion model, and update and optimize the model parameters of the diffusion model based on the measured data of the diffusion model;

[0184] The infrared image processing module is used to deploy the diffusion model obtained by updating and optimizing, and to perform infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result, so as to complete the optimization of the infrared image.

[0185] The working principle and beneficial effects of the above technical solution are: by iteratively superimposing non-uniform noise on the initial infrared image for a target number of times and recording the results of each superposition, the infrared image training set is accurately and effectively constructed; secondly, the infrared image training set is learned and trained by a deep neural network to accurately and reliably construct a diffusion model, and the diffusion model is continuously updated and optimized according to the test data of the diffusion model, thereby providing convenience and guarantee for the optimization of infrared images and ensuring the accuracy and reliability of the diffusion model finally obtained; finally, the obtained diffusion model is deployed, so as to facilitate the fast and effective processing of different infrared images to be processed according to the deployment results, thereby ensuring the processing effect of the infrared image while improving the processing efficiency, and retaining the detailed features of the infrared image while denoising the infrared image, effectively balancing the relationship between denoising and image details, and ensuring the optimization reliability of the infrared image.

[0186] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A deep learning-based infrared image optimization method, characterized in that: include: Perform a target number of non-uniform noise iterative superpositions on the initial infrared image, and independently record each non-uniform noise superposition result to construct an infrared image training set; Based on the deep neural network, deep learning is performed on the infrared image training set to build a diffusion model, and the model parameters of the diffusion model are updated and optimized based on the measured data of the diffusion model; wherein, based on the deep neural network, deep learning is performed on the infrared image training set to build a diffusion model, including: Acquire the infrared image training set, and separate the image contour and the infrared object image of each infrared image in the infrared image training set; Based on the deep learning network, the first target thread and the second target thread are respectively called according to the splitting result to process the image contour and the infrared object image according to the preset pixel gradient feature calculation kernel function to obtain the contour gradient feature and the infrared object gradient feature; Summarizing the contour gradient features and the infrared object gradient features to obtain the change rate of the infrared image in different directions, and determining the image gradient of the infrared image based on the change rate; Based on the image gradient, the balance parameters of denoising different types of infrared images and retaining infrared image details are determined, and the diffusion coefficients of denoising different types of infrared images are determined based on the balance parameters, and the basic equation of difference denoising is constructed based on the diffusion coefficient and the benchmark denoising requirements; At the same time, an application scenario of the denoised infrared image is determined, and an image enhancement dimension of the denoised infrared image is determined based on the application scenario; Determine the control variables and hybrid processing schemes of different image enhancement dimensions within the display range of the device based on the visual expression requirements, and formulate the image enhancement processing mechanism for infrared images based on the control variables and hybrid processing schemes; The basic equations of difference denoising and image enhancement processing mechanism are trained and a diffusion model is constructed based on the training structure. The diffusion model obtained by the update optimization is deployed, and the infrared image to be processed is subjected to infrared image denoising and enhancement processing based on the deployment result to complete the optimization of the infrared image, including: Based on the deployment results, the diffusion model is multi-dimensionally connected with the working platform, and based on the multi-dimensional connection results, infrared images to be processed uploaded by different working platforms are received; The infrared images to be processed are sorted based on the upload timestamps, and the infrared images to be processed are sequentially transmitted to the diffusion model based on the sorting results, and the infrared images to be processed uploaded at the same time are subjected to parallel infrared image denoising and enhancement processing in the deployed diffusion model; The parallel infrared image denoising and enhancement processing results are fed back to each working platform in turn to complete the optimization of the infrared image.

2. The infrared image optimization method based on deep learning according to claim 1, characterized in that: Before performing the target number of non-uniform noise iterative superposition on the initial infrared image, the following steps are included: Access the image library based on the server, and retrieve the record log of the image library based on the access result; Parse the record logs to determine the basic storage parameters of the infrared images in the image library, and determine the extraction scheme for the infrared images based on the basic storage parameters; The infrared image in the image library is extracted based on the extraction scheme, and the initial infrared image is obtained based on the extraction result.

3. The infrared image optimization method based on deep learning according to claim 1, characterized in that: Perform a target number of non-uniform noise iterative superpositions on the initial infrared image, including: Acquire the obtained initial infrared image, and determine a difference noise superposition scheme for different initial infrared images based on image features of the initial infrared image; Determine available noise of different initial infrared images based on a difference noise superposition scheme, and generate noise images of the same size according to size parameters of the initial infrared images based on the available noise; Based on the noise superposition requirement, the weight parameter of the noise image when iteratively superimposing the non-uniform noise with the initial infrared image is determined; The noise image is adaptively enhanced based on the weight parameter, and the adaptively enhanced noise image is iteratively superimposed with the initial infrared image for a target number of non-uniform noises.

4. The infrared image optimization method based on deep learning according to claim 3 is characterized in that: The noise image after adaptive enhancement processing is iterated with the initial infrared image for a target number of non-uniform noise superpositions, including: Performing foreground and background separation on the initial infrared image, and determining image adjustment variable values ​​for the initial infrared image based on the foreground and background separation result; Determining the retention features of the initial infrared image and the image contrast based on the image adjustment variable values; The noise image and the initial infrared image are converted into an image matrix based on the preserved features and image contrast, and the noise image and the initial infrared image are added based on the image matrix to complete the iterative superposition of non-uniform noise.

5. The infrared image optimization method based on deep learning according to claim 1, characterized in that: Each non-uniform noise superposition result is recorded independently to construct an infrared image training set, including: An image version monitoring thread is constructed based on the state change parameters of the initial infrared image, and the non-uniform noise superposition result of the initial infrared image is fully monitored based on the image version monitoring thread; Construct a record table, and determine the basic parameters corresponding to each non-uniform noise superposition result and the processed infrared image corresponding to each non-uniform noise superposition result based on the full monitoring results, and record the basic parameters in sequence in the record table; The processed infrared image corresponding to each non-uniform noise superposition result is converted into a separate file, and the separate file is associated with the corresponding basic parameters in the record table to obtain an infrared image training set.

6. The infrared image optimization method based on deep learning according to claim 1, characterized in that: The model parameters of the diffusion model are updated and optimized based on the measured data of the diffusion model, including: Acquire multiple test infrared images, and sequentially input the multiple test infrared images into the diffusion model for processing to obtain measured data of each test infrared image; Retrieve the reference infrared optimized image of each test infrared image, and perform image key point feature similarity matching between the reference infrared optimized image and the measured data; Determine the difference ratio between the measured data and the reference infrared optimized image based on the matching result, and when the difference ratio exceeds a preset threshold, determine the updated optimization dimension of the diffusion model and the updated optimization parameter on the updated optimization dimension based on the measured data; The model parameters of the diffusion model are updated and optimized based on the update optimization dimension and the update optimization parameter on the update optimization dimension.

7. The infrared image optimization method based on deep learning according to claim 1, characterized in that: Deploy the updated and optimized diffusion model, including: Determine the deployment environment of the diffusion model based on the usage scenario of the diffusion model, and determine the environmental adaptation parameters for the deployment environment in the server based on the operation requirements of the diffusion model; Adapt the deployment environment based on the environment adaptation parameters, and export the diffusion model based on the environment adaptation result to obtain a diffusion model file; Determine the current available space in the server based on the basic operating parameters of the server, and adaptively adjust the file size of the diffusion model file based on the current available space to obtain a deployable diffusion model file; Based on the environmental characteristics of the deployment environment, the low-code is logically combined to build the model deployment code, and the model deployment code and the deployable diffusion model file are uploaded to the deployment environment; Based on the preset deployment service, the deployable diffusion model is deployed according to the uploaded result, and the interface test is performed on the deployment result. When the interface communication status meets the preset requirements, the deployment of the diffusion model is completed.

8. An infrared image optimization system based on deep learning, characterized in that: include: The training set construction module is used to iteratively superimpose non-uniform noise on the initial infrared image for a target number of times, and independently record each non-uniform noise superposition result to construct an infrared image training set; The model building and optimization module is used to perform deep learning on the infrared image training set based on a deep neural network, build a diffusion model, and update and optimize the model parameters of the diffusion model based on the measured data of the diffusion model; wherein, performing deep learning on the infrared image training set based on a deep neural network to build a diffusion model includes: Acquire the infrared image training set, and separate the image contour and the infrared object image of each infrared image in the infrared image training set; Based on the deep learning network, the first target thread and the second target thread are respectively called according to the splitting result to process the image contour and the infrared object image according to the preset pixel gradient feature calculation kernel function to obtain the contour gradient feature and the infrared object gradient feature; Summarizing the contour gradient features and the infrared object gradient features to obtain the change rate of the infrared image in different directions, and determining the image gradient of the infrared image based on the change rate; Based on the image gradient, the balance parameters of denoising different types of infrared images and retaining infrared image details are determined, and the diffusion coefficients of denoising different types of infrared images are determined based on the balance parameters, and the basic equation of difference denoising is constructed based on the diffusion coefficient and the benchmark denoising requirements; At the same time, an application scenario of the denoised infrared image is determined, and an image enhancement dimension of the denoised infrared image is determined based on the application scenario; Determine the control variables and hybrid processing schemes of different image enhancement dimensions within the display range of the device based on the visual expression requirements, and formulate the image enhancement processing mechanism for infrared images based on the control variables and hybrid processing schemes; The basic equations of difference denoising and image enhancement processing mechanism are trained and a diffusion model is constructed based on the training structure. The infrared image processing module is used to deploy the diffusion model obtained by updating and optimizing, and perform infrared image denoising and enhancement processing on the infrared image to be processed based on the deployment result to complete the optimization of the infrared image, including: Based on the deployment results, the diffusion model is multi-dimensionally connected with the working platform, and based on the multi-dimensional connection results, infrared images to be processed uploaded by different working platforms are received; The infrared images to be processed are sorted based on the upload timestamps, and the infrared images to be processed are sequentially transmitted to the diffusion model based on the sorting results, and the infrared images to be processed uploaded at the same time are subjected to parallel infrared image denoising and enhancement processing in the deployed diffusion model; The parallel infrared image denoising and enhancement processing results are fed back to each working platform in turn to complete the optimization of the infrared image.

Images